United States
Environmental Protection
Agency
Atmospheric Research and
Exposure Assessment Laboratory
Research Triangle Park NC 27711
Research and Development
EPA/600/S3-90/046 July 1990
&EPA Project Summary
Evaluation of the Regional
Oxidant Model (Version 2.1)
Using Ambient and Diagnostic
Simulations
Thomas E. Pierce, Kenneth L. Schere, Dennis C. Doll, and Warren E. Heilman
This Project Summary discusses
an evaluation of the latest version of
EPA's Regional Oxidant Model,
ROM2.1. In the ambient evaluation,
model estimates were compared with
ambient measurements of hourly sur-
face ozone collected on 26 days
during the summer of 1985 In the
northeastern United States. Observed
and modeled maximum daytime con-
centrations agreed, on average, to
within 2 ppb (79 ppb versus 77 ppb).
The model tended to underestimate
at the higher extremes of the fre-
quency distribution. The 95th-percen-
tile value was underestimated by 8
ppb (127 ppb versus 119 ppb). Under-
estimates at the upper percentiles
were more prevalent in the southern
and western portions of the model
domain. Estimated and observed spa-
tial patterns of three day maximum
ozone generally showed good agree-
ment. ROM2.1 improved noticeably
over ROM2.0 with regard to the orien-
tation of the high-ozone plumes in
the Northeast Corridor. A unique as-
pect of the ambient evaluation was
an assessment of the model's ability
to estimate boundary conditions for
the Urban Airshed Model. Near New
York City, estimated and observed
boundary conditions agreed to within
4 ppb (57 ppb versus 61 ppb). Model
performance was degraded, however,
during some situations with dynamic
mesoscale wind flow conditions.
ROM2.1 also underwent a series of
diagnostic tests to investigate the
accuracy of its numerical solution
algorithms. When the model was
subjected to extremely steep concen-
tration gradients (steeper than those
observed in the ambient atmo-
sphere), the model did not conserve
mass during a 48 h simulation, devia-
ting by as much as 18% from the
initialized value. However, tests with a
mass-corrected version of the full
simulation model showed that predic-
ted ozone values deviated only slight-
ly (less than 4%) from the original
model.
This Project Summary was
developed by EPA's Atmospheric
Research and Exposure Assessment
Laboratory, Research Triangle Park,
NC, to announce key findings of the
research project that is fully
documented in a separate report of
the same title (see Project Report
ordering information at back):,
Introduction
After realizing that summertime epi-
sodes of photochemical smog, were a
regional and not a localized urban phen-
omenon, EPA embarked on a research
program in the middle 1970s to develop
a regional-scale computer model for sim-
ulating the transport and fate of ozone
(O3) and its precursors. In 1983, this work
resulted in the first operational version of
the EPA Regional Oxidant Model
(ROM1). Testing of ROM1 and analysis
of field data prompted additional
improvements, including the ability to
model biogenic emissions, horizontally-
varying layer thicknesses, and improved
deposition relationships. After a period of
development, the first application version
-------
of the ROM (Version 2.0) was completed
in 1986. It has since been used for
several studies in the Northeast and the
Gulf Coast regions of the U.S., and it has
undergone an intensive evaluation using
field data collected in 1980.
Additional needs by the northeastern
slates and the emergence of new model
improvements prompted the develop-
ment of ROM2.1. Improvements to
ROM2.1 over ROM2.0 included an
expanded modeling domain, an updated
methodology for computing biogenic
emissions of hydrocarbons, revisions in
the objective wind field interpolator to
remove known biases in the low-level
wind flow, and a more sophisticated
anthropogenic emissions processor.
Creation of ROM2.1 was primarily
motivated by the Regional Ozone Model-
ing for Northeast Transport (ROMNET)
project. In general, the objective of
ROMNET is to investigate interurban
transport of ozone and its precursor
emissions in the Northeast U.S. to sup-
port state air pollution control agencies in
the development of State Implementation
Plans to achieve the National Ambient Air
Quality Standard (NAAQS) for 03. In this
regard, ROM is being used to estimate
regional O3 concentrations and to
examine the effectiveness of a variety of
hydrocarbon (VOC) and nitrogen oxides
(NOx) emissions control strategies for
reducing O3. Another major objective of
the ROMNET project is to use ROM
calculations for deriving boundary condi-
tions for urban-scale applications with the
Urban Airshed Model (UAM). Because of
ROM's importance for comparing the
estimated effectiveness of emission con-
trol strategies and for estimating boun-
dary conditions, representatives of the
ROMNET project asked EPA's Office of
Research and Development to evaluate
the performance of ROM2.1.
The approach we took complements
the evaluation of ROM, Version 2.0. The
primary difference between the two
evaluations is that the ROM2.0 evaluation
used a special field-study data set from
1980, while our evaluation used routinely-
collected data from 1985. Our evaluation,
unfortunately, could not be as compre-
hensive as the ROM2.0 evaluation, which
had access to extensive field measure-
ment data for ozone, hydrocarbons, and
nitrogen oxides as well as aircraft
transects. We were able to use only
routine data stored in EPA's Aerometric'
and Information Retrieval System (AIRS).
Therefore, the ambient evaluation of
ROM2.1 was limited to hourly observa-
tions of surface ozone from state and
local agency monitoring sites.
In this project, we compared observed
and modeled ozone concentrations for
selected periods of high ozone observed
during the summer of 1985. Periods from
1985 were chosen because they
correspond to the base year emissions
inventory. The objectives of the ambient
evaluation were (1) to examine overall
evaluation statistics to determine whether
a general bias exists in the model
calculations, (2) to look at spatial patterns
of maximum concentration to determine
whether a spatial bias exists, and (3) to
examine the model's applicability for
determining UAM boundary conditions.
In addition to the ambient evaluation,
we performed a series of "stressful"
diagnostic tests on the model. In the
original development of the first genera-
tion ROM (ROM1), a number of
diagnostic tests were performed to probe
the accuracy of the numerical algorithms
used to solve the equations that sim-
ulated physical and chemical processes.
These tests were designed in a hier-
archial fashion beginning with a chem-
istry-only simulation. Transport, was
added to the chemical simulation, then
vertical .mixing, and finally source emis-
sions. These tests represented the, next
step after independent (external to the
ROM framework) tests of the model's
numerical algorithms. Results from this
original set of diagnostic tests, demon-
strated that the model faithfully 'repre-
sented solutions to the relevant
equations.
Since the time, of ROM1, numerous
changes have been made to the entire
ROM modeling system culminating in .the
most recent version of the second-gen-
eration ROM model (ROM2.1). Although
there were significant changes' to the
chemical and physical processes simu-
lated within the ROM, few changes were
made to the basic numerical algorithms
employed to solve "these processes.
Nevertheless, a new round of diagnostic
testing was proposed for RO'M2.1 that
would complement the earlier work done
with ROM1. We performed the'se
diagnostic tests on the production version
of ROM2.1 used for the ambient
evaluation. In the diagnostic evaluation,
the accuracy of the numerical algorithms
was assessed by evaluating the model's
ability to conserve mass for .these
diagnostic tests.
Ambient Evaluation
Because we wanted the model
simulations to correspond to the base
year of the 1985 NAPAP emissions
inventory, we examined ozone monitoring
data from 1985 for candidate episodes. In
particular, we were interested in periods
when ozone exceedances (hourly
concentrations greater than 120 ppb)
were observed throughout the Northeast
Corridor. After identifying two candidate
episodes, we examined these episodes
further for possible starting and ending
dates for model simulations, recognizing
that the ROM is designed for three-day
segments starting at noon and that the
first segment should be initialized with
"clean" (low ozone/precursor concentra-
tions) conditions. The episodes selected
for modeling were July 7-22 and August
7-16.
Two types of databases were
assembled for the ambient evaluation:
hourly concentrations of ozone from (1)
model estimates and (2) observations. To
produce the modeled database, we
executed the ROM for the two episodes.
Model inputs included National Weather
Service surface and upper-air meteor-
ological data, observed ozone concentra-
tions for estimating boundary conditions,
and hydrocarbon and nitrogen oxide
emissions (both anthropogenic and
biogenic).
The ROM is a three-layer Eulerian
grid-scale model that estimates hourly
photochemical species concentrations for
a 64 by 52 grid. Each grid cell is 1/6°
latitude by 1/4° longitude, or approx-
imately 19 km by 19 km. For this study,
we evaluated only hourly ozone concen-
trations from layer 1, because concentra-
tions from this layer most closely repre-
sent surface ozone observations. In layer
1, an individual grid cell is typically ~100
m in vertical extent at night, and 200-500
m deep during the day.
We developed three different model
databases for the evaluation. (1) Point
estimates from gridded data: For the
portion of the evaluation concerned with
general statistics, we interpolated gridded
estimated ozone values to actual monitor-
ing locations using a biquintic interpola-
tion scheme that is consistent with the
method used in the model. (2) Contoured
values of gridded data: For analyzing
spatial patterns, we used an objective
contouring algorithm to produce
computer graphics depicting concentra-
tion fields based on the gridded ROM
data. (3) Interpolated values derived
using the ROM/UAM interface method:
For the portion of the evaluation
concerned with boundary conditions, we
employed a fairly elaborate interpolation
scheme, described in the project report,
that is consistent with the ROM/UAM
interface developed for the ROMNET
_
-------
program. This interpolation allowed us to
transform boundary conditions from
ROM's 19-km grid size to the grid size of
the UAM domain, typically 4 to 10 km.
Analogous to the model databases, we
developed three observation databases:
(1) a set of observations used for
developing overall statistics; (2) a set of
observations used for creating contour
plots; and (3) a set of observations used
for developing UAM boundary conditions.
We obtained ozone concentrations from
monitoring data archived in EPA's
Aerometric Information and Retrieval
System (AIRS), which contains a national
database of hourly ozone (O3) concentra-
tions and information on monitoring sites.
Hourly 03 concentrations were selected
for sites located in the U.S. portion of the
ROMNET domain for the two episodes
that we modeled with ROM. (Canadian
03 monitoring data are not included on
AIRS and were not readily available for
this analysis.) An extensive review and
screening of the data was performed. We
included only daytime values (0800 h
LST to 1900 h LSI) in the evaluation
because nighttime observations are
influenced by localized processes that
often include scavenging of O3 by NOX
emissions and therefore do not reflect
vertically-integrated 03 concentrations in
layer 1 of the ROM. Furthermore, we
excluded a site's data on days missing
more than 25% of their observations. We
also examined the data for extremely
high or low values. Several sites, such as
Poughkeepsie (NY), were eliminated
because mean daytime O3 concentra-
tions were consistently below 50 ppb and
may have reflected local NOX scaveng-
ing. Of the more than 200 sites in the
original database, 187 of these were used
in computing general statistics. For
portions of the analysis, the data were
divided into five geographical groups.
The monitoring data used for
developing the UAM boundary condition
database were given special consider-
ation. The approach we followed is
consistent with previous modeling studies
that used monitoring data to prescribe
UAM boundary conditions. The assign-
ment of sites to a boundary location
depended on the prevailing wind
direction for that day. If more than one
site was available for a location, the
hourly concentrations were averaged.
After averaging at six locations along the
UAM boundary, concentrations were
spatially interpolated (using linear
avearaging). To be consistent with the
ROM/UAM interface method, we then
used the hourly concentrations to create
three-hour running averages.
In the ambient evaluation portion of
the study, we found good overall
agreement. For a 26 day simulation,
mean concentrations of the modeled and
observed daily maximum concentrations
agreed to within 1%. Concentrations at
the higher ends of the frequency
distributions were slightly underestim-
ated; the 95th-percentile observed daily
maximum concentration was 127 ppb
while the estimated concentration was
119 ppb. The tendency to underestimate
peak concentrations is to be expected
with a coarse grid model such as the
ROM because of the spatial averaging
that occurs with Eulerian grid computa-
tions.
In the Northern Corridor and Southern
Corridor geographical groups (groups 1
and 2, respectively), model performance
was good, particularly in group 2. The
group 1 mean observed and modeled
daily maxima agreed to within 11 % and
the 95th-percentile observed and
modeled daily maxima agreed to within
5% (both values were overestimated). For
group 2, the mean daily maxima were
within 3% of each other and the 95th-
percentile values were within 7% (both
values were underestimated). The
quantile-quantile plots of observed and
modeled daytime hourly concentrations
showed the same kinds of tendencies:
overestimation in the upper quantiles of
group 1 and underestimation in the upper
quantiles of group 2, as well as better
overall agreement for group 2 than for
group 1 - only the top 15% of group 1 's
estimates were within 10% of the
observations, while for group 2 the top
70% of the estimates were within 10% of
the observations. The medians in the
time series plots of daily maxima for
these two groups showed analagous
underestimate-overestimate tendencies,
and these plots also showed that
exceedances (values over 120 ppb) were
overestimated in group 1 (ten versus
eight) and underestimated in group 2 (six
versus ten).
Model performance in group 3 (the
southwestern part of the domain -
southern Virginia, West Virginia, Ohio,
and western Pennsylvania), an area
removed from the extensive metropolitan
area of the Northeast Corridor, was
noticeably poorer than the performance
for groups 1 and 2. The group 3 mean
daily maximum was underestimated by
12% and the 95th-percentile daily
maximum was underestimated by 19%.
The quantile-quantile plot for group 3
showed that the top 35% of the values
were underestimated by more than 10%.
The group 3 time series plot showed that
12 out of 14 medians were under-
estimated, and that exceedances were
dramatically underestimated (six versus
zero). Underestimates of upper-quantile
concentrations in this group perhaps can
be attributed to the relatively small-scale
urban plumes and uncertainties in
estimating naturally-occurring emissions
of NOX and hydrocarbons. However,
deficiencies in anthropogenic emissions
inventories should not be ignored and
ongoing efforts to improve them should
continue.
Groups 4 and 5 (the northern part of
the domain, excluding the Northeast Cor-
ridor) had observations close to back-
ground values, and model estimates
generally showed good agreement with
observations, especially for the upper
values.
Spatial patterns of the three-day
maximum concentrations usually showed
reasonable model performance. The
modeled magnitude and orientation of
ozone plumes in the northern portion of
the Northeast Corridor, especially around
New York City, compared well with
observed plumes. The model also did an
excellent job of predicting high ozone
levels around coastal sections of Maine.
We believe that ROM2.1 performed
better than ROM2.0 in these areas
because ROM2.1 includes a correction
for the westerly bias that occurred in
ROM2.0's low-level wind flows. Dif-
ferences between observed and modeled
plumes were most evident during epi-
sodes experiencing coastal troughs and
squall lines. Underestimates in the Wash-
ington, DC, area that were reported in the
ROM2.0 evaluation were seen again in
our evaluation. In addition, ROM 2.1
tended to underpredict rural peak
concentrations of ozone by about 20 ppb.
Few monitoring data are available for
evaluating model performance in rural
areas, so it is hoped that measurements
being taken under the auspices of the
National Acid Precipitation and Assess-
ment Program (NAPAP) will aid in future
evaluation of ozone concentrations.
The most rigorous portion of this
analysis was an evaluation of the model's
ability to estimate boundary conditions
for UAM application. Although we do not
recommend applying the ROM in a
deterministic manner, some type of
reliable estimation scheme is needed for
prescribing boundary conditions when
the UAM is to be applied for future-year
emission control strategies. Also, model
estimates are needed because measure-
ments aloft are typically not available for
specifying the vertical structure of
pollutant concentrations. Furthermore, our
-------
analysis showed that using monitoring
data to estimate boundary conditions
requires a great deal of subjectivity and
is fraught with uncertainty.
Overall, ROM2.1 estimates compared
quite well with the monitoring data for
estimating DAM near-surface boundary
conditions in the OMNYMAP domain. The
model overestimated the mean concen-
tration for all daytime hours by just 5 ppb
(8%) and underestimated the 95th-
percentile value by 9 ppb (9%). However,
we saw significant day-to-day variability
in model performance, ranging from a
maximum underestimate of 24% to a
maximum overestimate of 49%. A case
study performed for July 10, 1985,
demonstrated that small-scale meteoro-
logical features can cause dramatic
effects on model performance, apparently
because such features are not captured
in the ROM's overall interpolation of
meteorological data. We found that a
squall line resulted in a significant
overestimate in ozone concentrations
along UAM's western boundary (the in-
flow boundary on that day). We therefore
caution UAM users to carefully review
ROM estimates and be aware of
mesoscale flow conditions. On most
days, however, ROM2.1 did a reasonable
job estimating UAM boundary conditions.
Diagnostic Tests
In order to perform diagnostic tests on
ROM2.1 and its most critical processors
under known atmospheric conditions, a
specialized processor system has been
developed to allow the specification of
certain wind field and model layer
patterns in an analytically defined atmo-
sphere. By predetermining the wind field
and model layer depth patterns in time
and three-dimensional space, the capac-
ity of the ROM and specific processors to
obey various physical laws may be
tested.
Particular wind flow and layer defini-
tion sets for each test case are defined.
Also defined are the wind field and the
vertical layer boundaries, each of which
may be dependent on the various
physical quantities that are analytically
set in an earlier meteorological proces-
sor. Eight diagnostic test cases are de-
fined. Horizontal diffusion due to turbulent
eddies is set to zero for these tests. The
temperature lapse rate is defined to be
adiabatic, which is a good approximation
for the actual atmospheric mixed layer
where the ROM is designed to work. The
only atmospheric definitions which do not
approximate the real atmosphere are the
neglect of the effect of water vapor on the
specific gas constant for air and the
assumption of no horizontal diffusion due
to turbulent motion. The neglect of water
vapor effects is necessary to achieve a
tractable solution of the atmospheric
definition equations.
Test case 1A is designed to test the
horizontal transport and interlayer mass
flux mechanisms in the model. For this
test case, a circular anticyc'lonic wind
field is specified identically for all three
model layers. A temporally oscillating
pattern of ROM layer depths was used to
provide the mechanism for vertical mass
flux between ROM layers. The three layer
tops are kept horizontally flat at all times,
with the heights of the top of layer one
and two oscillating up and down in time.
The depth of the full model domain is
fixed over all space and time (1500 m).
For test case 1A, all chemical reactions
are turned off in ROM2.1. Only one
chemical species, used as a mass tracer,
is given a spatially-dependent initial con-
centration that differs from the prescribed
background concentration. This species
is given a conical-shaped initial concen-
tration distribution. It should be noted that
the initial concentration magnitudes are
completely arbitrary, because the chem-
ical reactions are turned off for this test
case.
Results of this test show an oscillation
in the mass field over time. The period-
icity corresponds to that of the oscillation
in the ROM layer depths for this test.
Mass increases as much as 6% in the
model domain during the first third of the
simulation. Later, a mass decrease of
around 2% is evident. The overall trend
in the mass field is toward a decrease
over time. The degree of mass change is
not strong here, although the test does
suggest that mass conservation errors
may occur when the model layer
interfaces change significantly over time.
Test case 1B is very similar to 1A,
except that the layer heights vary over
space instead of over time. This case is
also designed to test the transport and
numerical algorithms of the model. The
wind field and the initial concentration
fields for the three ROM layers are the
same for test 1B as they were for 1A.
Results for test 1B show that the mass
total remains within close proximity of the
original mass amount during the simula-
tion. Maximum changes of about 4%
from the original mass are seen.
Test case 1C is designed to examine
the accuracy of the numerical transport
scheme alone, without vertical fluxes
occurring. The wind field and initial mass
fields for each ROM layer used here are
identical to those used for test case 1A.
The layer top heights are horizontally flat
and are constant in time for test 1C. This
allows for an examination of the ability of
the ROM to conserve mass during
periods of rotational (non-divergent) flow
with no vertical transport across the layer
interfaces. Results show that the total
mass within the domain grows by nearly
7% in the first third of the simulation, and
then stabilizes later at a mass increase of
approximately 8% over the initial mass.
Considering that there are no physical or
chemical sources or sinks of mass in the
model simulation, this is a significant
mass increase.
Test case 2 is a somewhat more
complex test of the numerical transport
algorithms in the ROM. For this test case,
a purely divergent (convergent) flow field
is used. The layer average wind fields for
each of the model layers are identical,
resulting in no vertical wind shear. A
west-to-east zonal wind field is defined
for each layer such that speed maxima
exist along the western and eastern
boundaries of the modeled region, and
speed minima exist along the central
longitude of the modeled region.
It was desired for test case 2 that the
, layer boundaries represent material sur-
faces, or surfaces across which there is
no flux of material. Thus, the layer
heights must be determined such that,
given the layer average wind fields and
the analytically defined physical vari-
ables, there is no transport of air across
the layer boundaries. This stipulation re-
quires that the layer heights be deter-
mined based on the defined wind field.
Results show that during the first half
of this simulation the total mass increases
by nearly 18% within the grid. This
corresponds to the region of convergent
winds and increasingly deep layer
heights. As the cloud mass enters the
region of divergent winds and increas-
ingly shallow layer heights the mass
increases level off and eventually the
mass begins to decrease. This decrease,
however, occurs at a much slower rate
than the earlier mass increase, thus
resulting in a net mass increase of
approximately 13% at the end of the
simulation.
In test 0, we attempt to simplify the
transport environment to isolate the
cause of the mass increase. The ROM
layer heights for test case 0 were
spatially and temporally constant over the
model domain and were set at the values
used in test case 1C. The wind field
prescribed was essentially constant in
space and time. The initial concentration
field for test 0 was similar to that
specified for test case 1 A.
-------
Results for this test show a total mass
increase of about 10% toward the end of
this simulation. This result is significant
because the test has isolated the numer-
ical errors to the horizontal transport
scheme, without the complicating effects
of vertical redistribution of mass.
To determine the degree to which the
noted mass increases from the numerical
advection algorithms were a function of
the magnitude of initial concentration gra-
dients, a second test (test OA) was per-
formed in which the concentration
gradient was diminished by nearly a
factor of four.
Results from test OB show that mass
is essentially conserved during the sim-
ulation, with a mass increase of only
0.6% at the end of the simulation. The
contrast between the results of this test
and test 0 demonstrate that the numerical
artifact of the mass increase is a function
of the magnitude in concentration
gradient.
In the final test case (OB) presented
here, we investigate the effect of "clip-
ping" negative concentration predictions
from the model run. Small negative
ripples of concentration near the edges of
large concentration gradients are pro-
duced when these gradients are advec-
ted. The negative concentrations are a
result of the limitations of finite difference
approximations to large sub-grid gradi-
ents. The negative values, in themselves,
do not pose a problem for the advection
routines. However, when the advection
solution is passed to the chemistry
portion of an air quality model, there
must be no negative concentrations since
these are not defined in a chemical
simulation. Negative values produced by
ROM's numerical transport solution are
"clipped", or set to 10-16. Since we are
not solving the chemical equations in
these analytical tests, it is possible to
retain any negative concentrations resul-
ting from the transport simulation. In test
case OB we have suppressed the neg-
ative clipping and allow the propagation
of any negative concentrations.
The layer depths are the same as
those used in test case 0, constant in
space and time. The wind field is also the
same as in test case 0, simple zonal flow.
The initial concentration field is the same
as that used in test case 2. Results
indicate an initial increase of mass of
about 0.5%, dropping back to about 0.4%
for the remainder of the simulation. This
value should be compared to the nearly
10% (and rising) normalized mass ratio at
the end of the test 0 simulation. It is
apparent that the clipping of negative
concentrations has introduced a signif-
icant mass increase during the advection
of sharp concentration gradients.
To test the effect of "clipping"
negatives in actual ambient ROM simula-
tions, a two-day simulation was per-
formed for July 6-7, 1988, a particularly
severe ozone episode in the Northeast
U.S. In addition to the base run, con-
taining any mass imbalances caused by
the "clipping" of negatives, a simulation
was performed in which a first-order cor-
rection was .made to the advected
concentration field to assure mass
conservation.
Results show that the differences
between the corrected and uncorrected
simulation results were almost always
less than 1%. Differences seen in the
NOX concentrations were greater than
those of the other species, but generally
under 10%. These results suggest that
the implementation of a mass-correcting
scheme in ROM's numerical advection
algorithm would be a desirable, although
probably not essential feature. We plan to
repeat the diagnostic tests discussed
here with the global mass-correction
algorithm in place and analyze the results
in detail.
Summary and
Recommendations
Using both an ambient evaluation and
a series of diagnostic tests, we evaluated
Version 2.1 of the Regional Oxidant
Model (ROM2.1). In the ambient
evaluation, we assessed ROM2.1's per-
formance for periods of high ozone in
July and August of 1985 in the north-
eastern U.S., using AIRS daytime hourly
surface ozone monitoring data. We com-
pared these observations with model
estimates in three types of analyses: (1) a
comparison of overall statistics to deter-
mine whether model estimates exhibited
a general bias, (2) a comparison of spa-
tial patterns of maximum concentrations
to look for spatial bias in the model
estimates, and (3) an assessment of the
model's applicability for determining
UAM boundary conditions. In the diag-
nostic tests, we assessed the accuracy of
numerical algorithms by evaluating the
model's ability to conserve mass; we per-
formed five tests that involved only the
horizontal transport algorithm and two
that involved both horizontal transport
and vertical flux.
In the ambient evaluation, model
estimates were compared with ambient
measurements of hourly surface ozone
collected on 26 days during the summer
of 1985 in the northeastern United States.
Observed and modeled maximum day-
time concentrations agreed, on average,
to within 2 ppb or 1.4% (79 ppb versus
77 ppb). The model tended to
underestimate at the higher extremes of
the frequency distribution. The 95th-
percentile value was underestimated by 8
ppb or 6.6% (127 ppb versus 119 ppb),
and the overall maximum value was
underestimated by 50 ppb or 22.7% (219
ppb versus 169 ppb). Underestimates at
the upper percentiles were more prev-
alent in the southern and western
portions of the model domain. Concentra-
tions at the lower end of the frequency
distribution were slightly overestimated.
Estimated and observed spatial patterns
of three day maximum ozone generally
showed good agreement. ROM2.1
improved noticeably over ROM2.0 with
regard to the orientation of the high-
ozone plumes in the Northeast Corridor
and the depiction of high concentrations
along the coast of Maine. Similar to
ROM2.0, a tendency to underestimate
peak concentrations near Washington,
DC was again evident with ROM2.1. A
unique aspect of the ambient evaluation
was an assessment of the model's ability
to estimate boundary conditions for the
Urban Airshed Model. Near the New York
City metropolitan area, estimated and
observed boundary conditions agreed to
within 4 ppb or 7.6% (57 ppb versus 61
ppb). Model performance was degraded,
however, during some situations with
dynamic mesoscale wind flow conditions.
In the second part of our evaluation,
we employed a series of diagnostic tests
to assess the model's ability to conserve
mass. For test cases 1A and 1B, in which
mass flux was allowed between layers,
mass changes were expected to occur in
individual model layers. For the other
tests, however, no change in mass was
expected for individual layers. For all
tests, there should have been no change
in mass for all layers combined (total
domain).
For most test cases, there were
serious departures from mass conserva-
tion. Only in test case OA (where the
concentration gradients were consider-
ably relaxed) and test case OB (where
negative clipping was suppressed) was
the total domain mass effectively
conserved. The results for the other five
test cases suggest a potential problem
with the ROM's numerical transport
procedures, despite earlier design tests
performed during the ROM's develop-
mental stages that showed no problems
with mass conservation. We have de-
layed further diagnostic tests, including
examining test case 3 with chemical
simulation, and further analysis, including
preservation of peak concentrations, until
-------
the mass conservation problem has been
corrected.
A first-order correction algorithm has
boon developed based on a global
assessment of the mass imbalance due
to the clipping of negative concentrations.
This algorithm was implemented and
tested on a two-day ambient simulation
during a high 03 concentration period.
Results have shown that the differences
between the corrected and uncorrected
simulation results were almost always
less than 1%. Differences seen in the
NOX concentrations were greater than
those of the other species, but generally
under 10%. These results suggest that
the implementation of a mass-correcting
scheme in ROM's numerical advection
algorithm would be a desirable, although
probably not essential feature. We plan to
perform further tests to assess the
degradation in computation time with the
inclusion of the mass correction scheme.
With this additional information, we will
weigh the improvements in accuracy of
the transport solver with the increases in
computation time. We will also repeat the
diagnostic tests discussed in this section
with the global mass-correction algorithm
in place and analyze the results in detail.
Our analysis has demonstrated the value
of this type of diagnostic testing in model
evaluation.
Our evaluation has suggested that
further improvements to the ROM are
warranted. We are improving the
specification of layer thicknesses and the
computation of naturally-occurring emis-
sions. In future years, we hope that a
dynamic meteorological processor can
be incorporated that will simulate non-
steady state flows. To continue making
advances in model development, addi-
tional monitoring data are needed for
examining other chemical species (such
as NOX, isoprene, formaldehyde, and
HNOa) and for fully evaluating model
performance in rural areas.
-------
-------
The EPA authors, Thomas E. Pierce, Kenneth L Schere, (also the EPA Project
Officer, see below), and Dennis C. Doll are on assignment from the
National Oceanic and Atmospheric Administration. Messrs. Pierce and
Schere are on assignment to the Atmospheric Research and Exposure
Assessment Laboratory, Research Triangle Park, NC 27711. Warren E.
Heilman is now with the U.S. Forest Service, East Lansing, Ml.
The complete report, entitled "Evaluation of the Regional Oxidant Model (Version
2.1) Using Ambient and Diagnostic Simulations," (Order No. PB 90-225
293IAS; Cost: $23.00, subject to change) will be available only from:
National Technical Information Service
5285 Port Royal Road
Springfield, VA 22161
Telephone: 703-487-4650
The EPA Project Officer can be contacted at:
Atmospheric Research and Exposure Assessment Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711'
United States
Environmental Protection
Agency
Center for Environmental Research
Information
Cincinnati OH 45268
Official Business
Penalty for Private Use $300
EPA/600/S3-90/046
------- |