United States
Environmental Protection
Agency
Atmospheric Sciences Research
Laboratory
Research Triangle Park NC 27711
Research and Development
EPA/600/S3-86/032 Sept. 1986
Project Summary
EPA Regional Oxidant Model:
ROM1 Evaluation for 3-4
August 1979
Kenneth L. Schere
The first generation U.S. Environ-
mental Protection Agency Regional Ox-
idant Model (ROM1) was tatted and
evaluated for Oj predictions on a two-
day test case episode in the northeast
U.S. during 3-4 August 1979. The period
was characterized by relatively high Os
concentrations in the southern Great
Lakes area where dear skies prevailed.
The highest observed hour-average Os
level monitored at a surface she during
the period was 159 ppb. The ROM1 in-
corporates more simplifying assump-
tions and algorithms than the second
generation version, ROM2, which is
now in preliminary testing stages and
will eventually become the production
version of the model. Evaluation results
for this test episode showed that the
ROM1 had approximately a 6% average
underpradtetion of 03 when all hours
and surface monitoring sites were con-
sidered. When the data were restricted
to only those observed and predicted
pairs of 0) values greater than 50 ppb
the average performance of ROM1 im-
proved to a 1% underpredktion. The
evaluation phase concerned with esti-
mating maximum dally 03 values
showed an 8% average underprediction
of the maximum value for the restricted
data subset. An analysis of individual 0,
plumes during the episode showed that
the average model performance for pre-
dicting the plume maximum concentra-
tion level ranged from 22% underpre-
diction to 38% overpredtetion.
The Project Summery we* developed
by EPA'* Atmospheric Sc/ences f?e-
search Laboratory, Retesrch Trlengle
Park, NC, to announce Irey findings of
the research pro/act mat t* hilly docu-
mented In a separate report of the same
fftfe (see Project Report ordering Infor-
mation at beck).
Introduction
The United States Environmental Pro-
tection Agency's Regional Oxidant
Model (ROM) has been under develop-
ment and testing for a number of years.
The ROM attempts to simulate all of the
important physical and chemical proc-
esses that affect the generation and dis-
persion of photochemical smog on re-
gional scales (up to 1000 km). In the
present configuration of the ROM do-
main, the model's horizontal resolution
is approximately 18.5 km. In the vertical
there are three and one-half layers. The
first and second layers encompass most
of the planetary boundary layer above
the surface of the earth, and the third
layer is the cloud layer extending from
cloud base to near the tops of any
cumulus-type clouds. Layer 0 is a shal-
low layer adjacent to the surface de-
signed to treat sub-grid scale effects in
the ROM. It is handled in a diagnostic
manner within the model.
The model domain used in the first
ROM application in the Northeast U.S.
includes areas covered by special field
monitoring projects during the sum-
mers of 1979 and 1980. During these
projects highly resolved air quality, me-
teorological, and source emissions data
bases were developed for the study
area, hereafter referred to as the NEROS
region (Northeast Regional Oxidant
Study). The extent of the ROM1 domain
in the NEROS application is 60 cells in
the East-West direction and 42 cells in
the North-South direction. The dimen-
-------
sions of each individual grid cell are 15
minutes longitude (E-W) and 10 min-
utes latitude (N-S), or about 18.5 km2
cells. The specific episode studied here
occurred during 3-4 August 1979, dur-
ing the first NEROS field project.
The ROM version used in this model-
ing study was the first generation model
(ROM1). Several aspects of the ROM1
modeling system are in preliminary
form as compared to the planned pro-
duction version of the model, the sec-
ond generation ROM2. One major dif-
ference between ROM1 and ROM2 is in
the chemical kinetic mechanism. ROM1
contains a 23-species, 36-reaction-step
mechanism developed by Dr. Ken De-
merjian of EPA, whereas ROM2 will con-
tain a 28-species, 70-reaction-step
mechanism. Carbon Bond IV (CB-IV),
developed by Dr. Gary Whitten of Sys-
tems Applications, Inc. The CB-IV mech-
anism contains an explicit reaction
pathway for isoprene, a biogenic hydro-
carbon species. This will allow the
ROM2 to accommodate biogenic spe-
cies in the source emissions inventory;
ROM1 has no provision for biogenic
species.
The data base used here to test and
evaluate the 03 simulation results from
ROM1 represents that of 3-4 August
1979. An assessment of the meteoro-
logical aspects of the data base indi-
cated that the impact of the synoptic
meteorological features on the photo-
chemical smog potential over the
NEROS study area were at a maximum
from the Ohio Valley through the Great
Lakes into Ontario for this two-day pe-
riod. Wind speeds were light but of per-
sistent direction in this area, and skies
were clear. The eastern seaboard cities
were under partially clear or overcast
skies during the period. Surface wind
directions were quite variable in the
East also. The effects of these meteoro-
logical conditions were probably to sup-
press the photochemical potential in the
East, although for brief periods during
the daytime when skies cleared there
would be a local increase in the poten-
tial.
The 3-4 August 1979 period was not a
major episode for 03, although values
exceeded 100 ppb in selected areas. On
3 August, despite the cloud cover over
the mid-Atlantic seaboard, some 03 val-
ues reached between 100 and 115 ppb
in the Washington and Philadelphia
areas and between 120 and 130 ppb in
the New Jersey, New York, and Con-
necticut area, where there was less
cloud cover on that day. Measured 03
values of approximately 100 ppb were
found downwind of Detroit and
Toronto. Other measured values in the
vicinity of the Great Lakes and Ohio Val-
ley were not high, except near the
southeast shore of Lake Erie, where
Conneaut, Ohio recorded 159 ppb as the
daily maximum.
On 4 August, the overcast and rainy
conditions along the Atlantic coast from
New York northward did, in fact, seem
to suppress 03 generation. South of
New York, however, relatively high lev-
els were measured near the Philadel-
phia area (143 ppb) and Washington,
D.C. (139 ppb) where skies had cleared.
Several observation stations along the
south shore of Lake Erie reported maxi-
mum 03 levels from 120 to 135 ppb.
Most other areas of the NEROS region
reported low values, except for a few
observations near 100 ppb downwind
of Detroit and Toronto.
Numerous aircraft transects were
made over the NEROS region during the
3-4 August period. The temporal pro-
gression of flights from west to east re-
flects the Lagrangian sampling strategy
of attempting to follow an area of high
pollutant concentrations over the Mid-
west as it is transported east. Flights
outside of the Lagrangian sampling
were made as well. The aircraft data
showed evidence of an area of elevated
03 concentrations aloft in the boundary
layer to the south and east of the Lake
Erie shoreline and possibly over the lake
itself on 3 August. The timing and loca-
tion of this area of high 03 concentration
indicated that it was not generated on
3 August, but instead advected to the
area from another time and location,
quite possibly from Detroit and the
heavy industrialized areas to the west of
Lake Erie. A second 03 plume of urban-
scale, probably from Detroit, was seen
over western Ontario on 4 August. In
general, the aircraft sampling during
this two-day episode over the NEROS
region indicated widespread boundary
layer 03 levels of 70 to 80 ppb with some
areas slightly higher or lower than this
range at times.
Results
The ROM1 model was run for the 48-h
period of 3-4 August 1979 using data
from the NEROS and SAROAD data
bases. The model was initialized
throughout its entire domain with a set
of tropospheric background species
concentrations that were computation-
ally adjusted for chemical equilibrium.
The same set of concentrations was
used for both lateral inflow and top
boundary concentrations throughout
the model simulation period.
The evaluation of the ROM1 results
for 03 prediction was completed in sev-
eral stages. First we compared model
predictions, interpolated to the surface
SAROAD station locations, with ob-
served values over all hours and recep-
tor sites. Next, an analysis of the maxi-
mum concentrations at receptor
locations was made, and finally a thor-
ough examination of individual plumes
from large source areas within the do-
main was performed. We ignored the
first 6 hours of predictions from the
model because of the strong influence
of initial conditions during this period.
The analysis of surface 03 observa-
tions and model predictions over all re-
ceptor sites and all hours showed that
the diurnal range in 03 concentrations
was considerably greater for the obser-
vations than for the model predictions,
where an average variation of only 10 to
15 ppb was shown. Bias values, or ob-
served minus predicted concentration
values, were generally negative at night
and positive during the day. The abso-
lute values of bias (gross error) were all
consistently rather high, indicating that
the hours with average bias near zero
actually contained a wide variation of
individual values with opposite signs.
The average episodic value of observed
03 for all hours at the surface monitoring
sites was 38.9 ppb, and the correspond-
ing model predictions for ROM1 layers 0
and 1 were 36.6 and 38.1 ppb, respec-
tively. These values represent a 5.9%
underprediction for layer 0 and a 2.1%
underprediction for layer 1. Since these
average 03 levels are near tropospheric
background values, the significance of
this analysis over all hourly values in
the simulation period was not particu-
larly great.
The next step in the evaluation was to
perform an hourly analysis with a sub-
set of the data values already used. The
criterion for choosing the subset was
based, in part, on the tropospheric back-
ground value of approximately 40 ppb
03 used in the ROMTs initial and
boundary conditions and the fact that
many of the surface monitoring site lo-
cations showed 03 predictions near this
value. Therefore, the data subset was
based on only those receptor points for
which the observed value and the level
0 and 1 predicted values of 03 were
-------
greater than 50 ppb. This should limit
the data to those sites most affected by
source emissions within the NEROS re-
gion. In fact, after this criterion was ap-
plied the number of sites used in the
analysis dropped from near 150 during
the midday hours to around 30, and
from around 120 during the nighttime
hours to 0.
The analysis of the data subset
showed that the average observed con-
centration at the receptor locations
varied from 51.0 ppb at 2000 h to 78.0
ppb at 1300 h on 3 August, and from
50.0 ppb at 0400 h and 0800 h to
80.7 ppb at 1600 h on 4 August. The cor-
responding predictions for layer 0
ranged from 56.0 ppb at 1900 h to 70.4
ppb at 1600 h on 3 August, and from
54.4 ppb at 0400 h to 76.2 ppb at 2000 h
on 4 August. The average episodic
value of observed Oa in the data subset
for all hours was 72.2 ppb, and the cor-
responding ROM 1 predictions for layers
0 and 1 were 68.9 and 71.2 ppb, respec-
tively. This represents an average 03
underpredictionof 4.6%for ROM1 layer
0 and 1.4% for layer 1, a slightly smaller
degree of underprediction than the val-
ues computed for the full data set. The
average values in the data subset were
however, significantly greater than the
tropospheric background 03 value.
The most significant parameter of in-
terest for the air quality policymaker
concerning ambient 03 is the daily
hourly average maximum value ob-
served at a monitoring station. The
model evaluation for maximum Oj con-
centration took place in two steps. First,
in the local maxima analysis, model pre-
dictions of maximum hourly average 03
levels at surface monitoring site loca-
tions were compared with measured
values at the sites. Second, in the global
maxima analysis, the concentration
magnitudes and position of plumes of
Oa within the NEROS region were com-
pared between the ROM1 predictions
and the observations at surface moni-
toring sites, to the extent that these sites
can define such plumes.
The local, daily maximum 03 concen-
tration, c(max,ob), at a measuring site
can be compared with the correspond-
ing model prediction in two ways. First,
the predicted maximum at the site,
c(max,pO) for ROM1 layer 0 or
c(max,p1) for ROM1 layer 1, can be
compared to c(max,ob). Second, a more
restrictive measure matches the predic-
tion that occurs at the same hour as the
observed maximum with the observed
concentration [c(pO,h-ob) vs. c(max,ob)
and c(p1,h-ob) vs. c(max,ob)].
A summary of average concentration
data from the local maximum values
analysis for the model simulation is pre-
sented in Table 1. Results from the full
data set, as well as the data subset
where all 03 concentrations are greater
than 50 ppb, are given. The number of
receptor stations dropped from 156 in
the full data set to 47 in the data subset.
Interestingly, the average observed
maximum value increased by only 2 to
4 ppb from the data set to the subset
because there were very few observed
maxima less than 50 ppb at the monitor-
ing site locations. Conversely, the aver-
age model predictions increased by 20
to 30 ppb from the data set to the subset
because of the large number of values
near the background 03 level. For the
looser pairing of observed and pre-
dicted maximum values, the average
bias at all receptor sites was 24.8 ppb on
3 August, representing a 34% underpre-
diction and 25.3 ppb on 4 August, repre-
senting a 33% underprediction for
model layer 0. For layer 1, the corre-
sponding values were 23.2 ppb (32%
underprediction) on 3 August, and 23.3
ppb (30% underprediction) on 4 August.
In the data subset where the lower 03
values are excluded, the results show a
marked improvement. On 3 August, the
average bias for layer 0 was 11.3 ppb
(15% underprediction) and on 4 August,
it was 5.6 ppb (7% underprediction). For
layer 1 the corresponding values were
9.2 ppb (12% underprediction) on 3 Au-
gust, and 2.7 ppb (3% underprediction)
on 4 August.
The global maximum values analysis
is performed to isolate for comparison
individual areas or plumes of high 03
concentrations from the model predic-
tions and the ambient observations.
The perspective of this analysis is
broader than that of the local maximum
values analysis in that model predic-
tions at the same location as the obser-
vation are not necessarily required. In-
stead, the eligible area from which the
model prediction is chosen is defined to
be the coherent region of concentra-
tions, or plume, from which the obser-
vation comes. This less restrictive pair-
ing permits us to match observations
and predictions based on supposedly
similar phenomenological events in the
physical processes producing the maxi-
mum Os values. This perspective also
permits us to include a larger area than
the single site location where the maxi-
mum value occurred.
In estimating model performance in
terms of the global maximum perspec-
tive, we made three comparisons of
model-predicted concentration with the
maximum observed concentration,
c(max,ob), in the Os plume. First, the
maximum observed concentration was
compared with the predicted maximum
at the same location as the monitoring
site of the observed maximum. Second,
the observed maximum was compared
with the predicted concentration at any
of the monitoring site locations in or
near the projected plume, and finally it
was compared with the maximum con-
centration found in the grid cell at the
center of the projected Oa plume. These
three values are designated (for ROM1
layer 1) as c(max,p1-A), c(max,p1-B),
and c(max,p1-C), respectively. These
comparisons range from the most re-
strictive to the least restrictive pairing of
concentrations in the global maxima
analysis.
Table 2 presents a summary of statis-
tics from the global analysis of 03 plume
Table 1.
Summary of ROM1 Results for Local Maximum 03 Values for 3-4 August 1379*
Full Data Set
Data Subset
(all concentrations > SO ppb)
79215
79216
79215
79216
Number of stations
Average c(max.ob)
Avenge c(max, pO)
Average c(pO,h-ob)
Average c(max,p1)
Avenge c(p1,h-ob)
156
73.2
48.4
40.0
50.0
41.6
154
77.8
52.5
43.5
54.5
45.1
47
75.5
64.2
60.0
66.3
62.0
48
81.8
76.2
68.9
79.1
71.4
'All concentrations are in ppb.
-------
Table 2. Summary of ROM1 Results for Global Maximum 03 Values for 3 - 4 August 1979*
Urban Plume Date c(max,ob) cfmax,p1-A) c(max,p1-B) c(max,p1-C)
Detroit
Toronto
New York
Detroit
Toronto
New York
Philadelphia
Average
S.D. .
3 Aug.
3 Aug.
3 Aug.
4 Aug.
4 Aug.
4 Aug.
4 Aug.
100
98
130
99
W1
144
132
115
20
90
110
97
88
66
59
101
87
18
122
110
97
94
107
88
130
107
15
158
144
117
162
227
114
135
151
38
*AII concentrations are in ppb.
maximum values. Seven individual
plumes were considered. The average
value of c(max,ob) recorded at surface
monitoring sites in these plumes was
114.9 ppb.* Three methods of pairing
this observed value with ROM1 predic-
tions were made. From the most restric-
tive to the least restrictive pairing the
corresponding layer 1 model predic-
tions averaged over the seven cases
were c(max,p1-A) = 87.3 ppb, c(max,p1-
B) = 106.9 ppb, and c(max,p1-
C) = 151.0 ppb. The last predicted value
is not restricted to measurement site lo-
cations, but can be at any grid cell
within the predicted plume area. This
measure has the greatest potential for
overproduction, and in fact we would ex-
pect this to be the case because the den-
sity of measurement networks is rarely
sufficient to capture the true 03 peak
value. The average bias for this pairing
was -36.1 ppb, implying an average
overprediction of about 31%. Note that
if we drop from consideration the
Toronto plume on 4 August, the aver-
age overprediction changes to about
23%. The very high predicted value over
Lake Ontario, which could not be veri-
fied since the area was out of the range
of the surface monitors, lends some
support for this action.
The average bias in the most restric-
tive pairing for which the model predic-
tion is interpolated to the same location
as the monitoring station was 27.6 ppb,
and the corresponding degree of under-
prediction was 21.6%. For the second
comparison, where the maximum
model prediction was restricted to mon-
itoring site locations in or near the 03
plume, but not necessarily at the site of
the observed maximum, the average
bias was 8.0 ppb with a corresponding
degree of underprediction of 4.4%.
The variance in the values of overpre-
diction or underprediction among the
seven plume cases considered was indi-
cated by the standard deviation (s.d.)
about the average. For the two most re-
strictive pairings, the s.d. was approxi-
mately the same in each case, 21 to
22%. For the least restrictive pairing, the
s.d. was approximately 54%, although
much of that variance is attributable to
the 4 August Toronto plume case. Al-
though layer 0 predicted values were
not considered in the global maximum
values analysis, the results would not
have been very different from those
shown for layer 1. The layer 0 predicted
maximum values were typically only a
few ppb less than the corresponding
layer 1 values.
Conclusions
The first generation Environmental
Protection Agency Regional Oxidant
Model (ROM1) has been evaluated for
03 concentrations using the NEROS data
base for the two-day test period, 3-4 Au-
gust 1979. The evaluation proceeded in
two stages. First, an overall analysis of
03 observations and predictions at re-
ceptor monitoring locations in the
model domain was made for all hours
of the simulation except the first six. In
the stage where all data were involved,
results showed that, on the average, the
ROM1 underpredicted 03 levels by 6% in
comparison with measurements at
monitoring sites. In a second part of the
first stage of analysis, the same manner
of evaluation was applied to a subset
of the data where only simultaneous 03
values above 50 ppb were included for
observed and predicted concentrations
at monitoring sites. In this case, the av-
erage ROM1 performance for the data
subset showed a 4.6% underprediction
for layer 0 and a 1.4% underprediction
for layer 1.
The second aspect of model evalua-
tion, using daily 03 maxima, also* pro-
ceeded in two steps. In the first step, the
daily maximum value at each receptor
location was compared to the predicted
maximum at that location as well as the
prediction made for the same time as
the observed maximum. For layer 1, the
ROM1 performance showed an average
31% underprediction for the first com-
parison and a 43% underprediction for
the second comparison. Results for
layer 0 were similar. When data were
restricted to those observed and pre-
dicted 03 pairs above 50 ppb, the model
performance for the local maximum
analysis improved by showing average
underpredictiqns of 8% and 15%, re-
spectively, for the above layer 1 com-
parisons. The improvement in model
predictions in this case was primarily
due to restricting the analysis to com-
parisons at monitoring stations loca-
tions where source emissions had a
larger effect on 03 levels than material
that was present initially and was over-
looked by the clean initialization proce-
dure.
The second step of evaluating the
ROMI's performance for 03 maxima
predictions involved an analysis of indi-
vidual plumes of 03 from major source
areas within the NEROS region. Seven
cases of such plumes were identified
and analyzed for the 3-4 August 1979
episode. The evaluation used three
types of comparisons of 03 maxima
ranging from a strict spatial pairing be-
tween observed and predicted values to
a fairly loose spatial pairing. Results av-
eraged over the seven plumes in the
test case showed a 22% underpredic-
tion for the strictest paired comparison,
a 4% underprediction for the intermedi-
ate comparison, and a 38% overpredic-
tion for the loosely paired comparison.
The uncertainty of these aspects leads
one to conclude that the results pre-
sented here are probably bounds on the
actual model performance. The ROM1
tended to perform better in those areas
of the domain where a well-organized
flow field was present and the assump-
tion of clean initial conditions was not
grossly violated. Applications with the
-------
second generation regional model,
ROM2, should be able to quantify the
model performance to a greater degree
than was done in this case.
The EPA author Kenneth L. Sehere (also the EPA Project Officer, see below) is
with the Atmospheric Sciences Research Laboratory. Research Triangle Park.
NC 27711.
The complete report, entitled "EPA Regional Oxidant Model: ROM1 Evaluation for
3-4 August 1979." (Order No. PB 86-216 886/AS; Cost $16.95. subject to
change) will be available only from:
National Technical Information Service
5285 Port Royal Road
Springfield. VA 22161
Telephone: 703-487-4650
7'he EPA Project Officer-can be contacted at:
Atmospheric Sciences Research Laboratory
U.S. Environmental Protection Agency
Research Triangle Park. NC27711
------- |