United States
Environmental Protection
Agency
Atmospheric Research and Exposure
Assessment Laboratory
Research Triangle Park NC 27711
Research and Development
EPA/600/S3-89/057 July 1989
Project Summary
EPA Regional Oxidant Model
(ROM2.0): Evaluation on 1980
NEROS Data Bases
Kenneth L. Schere and Richard A. Wayland
The second generation USEPA
Regional Oxidant Model (ROM2.0) has
been evaluated for the northeastern
United States using the 1980 NEROS
data bases. The theoretical basis of
the model and Its structure and
organization are described. The data
bases available from the summer
1980 period include routine air quality
and meteorological monitoring data
In addition to data from several
extensive field measurement projects
conducted during the summer of
1980 in the northeastern U.S. on
regional and urban scales. Also, a
complete emissions inventory, com-
posed of anthropogenic and biogenic
components, was assembled for the
1980 base year for use in air quality
modeling. The ROM2.0 evaluation was
conducted using quasi-deterministic
and diagnostic techniques. Strict
temporal and spatial pairing between
observations and predictions was not
used in the analysis. Model simu-
lation was conducted during the
period of July 12 to August 31, 1980.
Model performance over the simu-
lation period showed an overall 2%
overprediction of the daily surface
maximum O3 concentrations. ROM-
predicted concentrations, however,
had a narrower range for ambient O3
with underestimation of highest
values and overestimation of lowest
values. The spatial extent and con-
centrations of urban O3 plumes were
generally simulated well, although a
bias in the transport direction along
the East Coast caused frequent mis-
alignment of the plumes. Model
performance analyses using aircraft
data showed the model to under-
predict the regional O3 tropospherlc
burden under episodic conditions,
although individual plumes were
modeled well.
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
project that Is fully documented In a
separate report of the same title (see
Project Report ordering information at
back).
Introduction
The initial development of a regional
(-1000 km) air quality simulation model
began in the late 1970's after the
realization that photochemical smog often
extended beyond individual urban areas
to entire sections of the U.S. Interstate
transport of 03 and its precursors was
observed during field programs of the
1970's, especially in the Northeastern
U.S. Long-range transport of O3 and
precursors and multi-day chemical ef-
fects could not be properly treated by
existing urban-scale photochemical
models. The need became apparent for
an appropriate simulation model to test
the effectiveness of particular emissions
control strategies on O3 concentrations in
urban airsheds as well as region-wide.
The first generation EPA Regional
Oxidant Model (ROM 1.0) became opera-
tional in 1984. It was a test bed for the
future production version of the model,
the second generation ROM2.0. The
-------
earlier model contained a very con-
densed chemical kinetic mechanism,
could not treat natural hydrocarbons, had
limited treatment of vertical mass flux
induced by clouds, had constant layer
depths, and contained very limited terrain
effects. Nevertheless, ROM1.0 was used
for extensive testing of various emission
control scenarios in the Northeast U.S.
ROM2.0 became operational in 1987.
This second generation version of the
model was to become the production
version. It included a more sophisticated,
contemporary chemical kinetic mech-
anism capable of treating both anthro-
pogenic and natural precursor species.
The ROM2.0 system also corrected most
of the deficiencies and simplifications of
the first generation system, such as
cloud-induced mass flux, variable layer
depths, and terrain effects. Once opera-
tional, ROM2.0 was used for extensive
testing of emissions control scenarios
within the Northeast U.S. for EPA's Office
of Air Quality Planning and Standards
(OAQPS), as well as the Vice President's
Commission on Clean Coal Technology,
and the Congressional Office of Technol-
ogy Assessment.
A large field project in the Northeast
U.S. was planned and conducted con-
currently with the ROM model devel-
opment effort. The purposes of the field
program were to gather data to better
understand the important processes
responsible for photochemical smog on
the regional scale so they could be
properly simulated by the model and to
provide a data base for testing and
evaluation of the model. The field
program was carried out over two
summer periods during 1979 and 1980.
Measurements were taken from Ohio and
Michigan to the East Coast, both in
regional and in local projects. Extensive
aircraft measurements supplemented the
ground-based network. The data base
from 1979 was used primarily in the
model development, testing, and evalua-
tion phases of ROM1.0. The 1980 data
base was used almost exclusively for
model evaluation of ROM2.0.
This project presents the results of the
evaluation of ROM2.0 on the 1980
Northeast U.S. data base. Emphasis is
placed on the comparison between pre-
dicted and observed 03 concentrations,
although NOX and hydrocarbons are eval-
uated also, to the extent that the data
allow. Ground-based and aircraft data are
used in the analysis. The purpose of this
evaluation is essentially a confidence
building exercise in the ROM2.0 for its
use as an air quality planning tool.
Model Application
The ROM has been designed to
simulate most of the important chemical
and physical processes responsible for
the production of photochemically pro-
duced 03 on scales of 1000 km, or
several days of transport time. These
processes include horizontal transport,
atmospheric chemistry, nighttime wind
shear and turbulence episodes associ-
ated with the nocturnal jet, cumulus cloud
effects on vertical mass transport and
photochemical reaction rates, mesoscale
vertical motions induced by terrain and
the large scale flow, terrain effects on
advection, diffusion, and deposition, sub-
grid scale chemistry processes, emis-
sions of natural and anthropogenic pre-
cursors, and deposition. They are mathe-
matically simulated in the 3-D Eulerian
model with 3-1/2 vertical layers including
the boundary layer and the capping
inversion or cloud layer. Horizontal reso-
lution is 1/4° of longitude by 1/6° of
latitude, or about 18.5 km.
The particular application of ROM2.0
used in this evaluation exercise was
conducted on a historical data base, with
model simulation beginning at 1200 h,
local standard time (LST) on July 12,
1980 and continuing through 1200 h, LST
on August 31, 1980. The domain of
application is that of the Northeast U.S.,
shown in Figure 1. The simulation was
not reinitialized at any time after it began.
It was continuous in time, performed in
contiguous 3-day segments, with the
simulation results from the final step of
one segment used as restart conditions
for the initial step of the next segment.
The simulation period contained several
significant O3 episodes in the North-
eastern U.S., with measured concentra-
tions as high as 300 ppb.
The data bases used in this project
include meteorological, air quality, and
emissions components. The ROM
requires all three types of data for
simulating regional air quality. The
performance evaluation of the model pre-
dictions primarily requires the air quality
data base. The summer 1980 period was
chosen for this effort because it
coincided with several major field
projects, conducted in the Northeast U.S.,
designed to study the regional and urban
03 problems. These projects provided
special data bases which supplemented
the standard air quality and meteor-
ological measurements archived in EPA's
SAROAD (Storage and Retrieval of Aero-
metric Data) system and collected by
NOAA's National Weather Service
(NWS), respectively. The model h«
been designed to run in an operation
mode solely on these routinely collects
data bases. The special study data base
are used for model evaluation ar
research on model parameterizations.
The SAROAD data, supplemented t
monitoring data in southern Ontari
Canada obtained from Environme
Canada and the Ontario Ministry of tf
Environment, provided hourly measur
ments of O3, N02, and NOX at fixe
monitoring sites within the model domai
There were 214 sites where 03 mea
urements were made, 107 sites for NC
and 65 sites for NOX during the summ
1980 period. Most of the monitoring sit<
are within or near urban areas.
Hourly surface meteorological me
surements were available from ««2C
stations within the ROM domain in tf
NWS and Canadian meteorological ne
works. The ROM preprocessors assir
ilate raw meteorological data on atmo
pheric pressure, temperature, moistur
winds, and clouds from these location
In addition to the surface measurement
the North American upper air soundir
network contains 24 stations within ar
near the boundaries of the ROM doma
where twice daily upper air soundinc
provided measurements of pressur
temperature, moisture, and winds. Nir
of these stations are located within tf
ROM domain boundaries, and the fn
quency of soundings was increased
four times per day at these statior
during "intensive" NEROS field stuc
periods.
During the 1980 summer season EPA
Office of Research and Developme
sponsored two major field studies in tf
Northeast U.S. The first was the Nort
east Regional Oxidant Study (NEROS
The NEROS field measurements conce
trated on sampling strategies to clari
and parameterize essential process*
simulated within the ROM, to provic
input data for the model, and to provic
data with which to evaluate the mod(
The Persistent Elevated Pollutant Ep
sode (PEPE) study was performed
conjunction with the NEROS in 1980. I
focus was on a regional perspective co
cerning the broad regions of ha/
associated with large stagnant a
masses.
The Northeast Corridor Region
Modeling Program (NECRMP), spo
sored by EPA's OAQPS, included urb<
field studies during the summer of 19f
designed to collect the necessary i
quality and meteorological data nece
sary to apply the Urban Airshed Model
-------
N 38.00°
Figure 1. Northeast U.S. ROM domain. Each dot represents a corner of a grid cell.
Washington, DC, Baltimore, New York,
and Boston. The data requirements for
applying and evaluating this model led to
the development of a monitoring program
including air quality measurements by
aircraft upwind of the urban area in the
morning, and over and downwind of the
urban area in the afternoon.
There are two major components to the
emissions inventory data base needed by
the ROM system: the anthropogenic
emissions, and the biogenic emissions.
Anthropogenic emissions of NOX, CO,
and various categories of non-methane
hydrocarbons (NMHC) were obtained
from the 1980 National Acid Precipitation
Assessment Program (NAPAP) emissions
inventory. The final inventory for use by
ROM contains hourly emission rates on
the ROM's 18.5-km grid resolution for
NO, N02, CO, and the NMHC categories
of ethylene, olefins, paraffins, formalde-
hyde, higher aldehydes, toluene, xylene,
and non-reactives.
The biogenic portion of the emissions
inventory, developed at EPA, consists of
hourly, gridded values of natural hydro-
carbon species. Three basic components
are required to develop estimates of
hourly, gridded biogenic NMHC emis-
sions: (a) emission factors representative
of vegetative species indigenous to the
modeling region, (b) empirical relation-
ships between emission factors and
specific environmental parameters, and
(c) quantitative estimates of vegetation
density of the representative species in a
designated area. The development of the
biogenic inventory consisted of the
compilation and assimilation of each of
these factors.
To circumvent any model initialization
problems, the ROM is initialized on a
relatively clean day several days before
any periods of high 03 concentrations are
found to exist in the domain. The full
domain is assumed to contain spatially
invariant values of clean tropospheric
background concentrations of O3, NOX,
and NMHC.1 These species concentra-
tions were then allowed to chemically
equilibrate, using an independent chem-
ical solution module. The set of initial
values was spatially invariant in each
horizontal model layer. To further isolate
the model results from initial condition
artifacts, the results for the first 24 hours
of simulation are not used in the model
evaluation analysis.
Boundary condition problems cannot
be circumvented in the same manner as
the initial condition problems. The goal,
therefore, is to mitigate the effect of the
1 Clean tropospheric background values used are
O3 = 35 ppb, NOX = 2 ppb, NMHC = 15 ppb.
-------
boundary conditions. This has been done
to some extent by the specification of the
size of the model domain. While the area
of greatest interest in the simulations is in
the Northeast Corridor from Washington,
DC through Boston, the model domain
extends in an upwind direction to the
Ohio Valley and south to northern Virginia
and West Virginia. In this way the model
assimilates all of the significant upwind
sources potentially making an impact on
the Northeast Corridor and thereby
reducing the influence of boundary con-
ditions in that area. Obviously, the farther
west and south in the domain one goes
from the Corridor, the greater will be the
boundary condition influence.
We have assumed the same tropo-
spheric background conditions at the
ROM boundaries as were described for
the initial conditions, with one exception.
We allow for the fact that 03 may deviate
from this background value at lateral
boundaries. We set the boundary 03
concentration from ambient monitoring
data for each day of simulation. For a
given day and time period the same 03
value was used at all lateral boundaries.
Model Evaluation
Most methods of model evaluation
used in previous studies have been large-
ly deterministic. That is, the model con-
centration predictions for a specific
location are compared to observations
taken at that location on a given day. The
ROM's developer (R. Lamb) maintains
that there are inherent limitations on the
predictability of air quality models, partic-
ularly on regional and larger spatial
scales. He suggests that, even if the
model is formulated perfectly, the data
needed to drive the model are not
sufficient to exactly determine the state
of the atmosphere. This uncertainty in the
atmospheric state gives rise to cor-
responding uncertainties in the concen-
tration predictions from even a perfect
model. The magnitude of the uncer-
tainties in the concentration values is
directly proportional to the extent to
which concentrations at a receptor are
affected by distant sources. Furthermore,
the level of uncertainty increases with
increasing distance from the nearest
meteorological station and with increas-
ing distance to significant sources. These
ideas concerning the uncertainty in
atmospheric state translate directly into
uncertainty in the wind-driven transport
component of the air quality model
solution.
In the analyses used for this project we
use a quasi-deterministic mode for the
evaluation of ROM results when com-
paring predictions to surface-monitored
observations of O3, N02, and NOX. In the
quasi-deterministic mode we aggregate
concentrations at groups of receptor
locations and compare the aggregate
frequency distribution of concentrations
from the receptor group with that of the
observed concentrations from the group.
This is possible because there are suf-
ficient numbers of these monitoring
stations to form coherent groups for the
aggregation step in the analysis. For
surface monitoring of NMHC and also for
all aircraft monitoring there are not
sufficient stations, or, in the case of
aircraft, the data are obtained too
intermittently, to form groups for aggre-
gation. It is therefore more difficult to
implement the quasi-deterministic meth-
od of evaluation. In this case we form the
most appropriate spatial and/or temporal
averages of data to compare with
observations. With this combination we
attempt to maintain a balance of the use
of all available data with the conscious
desire to use the data appropriately. In
this context, the rigorous comparison of
data of dissimilar scales is considered
inappropriate.
Data Preparation
The evaluation of surface-based 03
concentrations follows along the lines of
the quasi-deterministic analysis. We
focus on the ability of the ROM to
simulate 03 concentration frequency
distributions at groups of receptor loca-
tions during the simulation period. Each
group of receptors shall have certain
characteristics common to all members
of the group. In our case we have chosen
the observed frequency distribution of 03
concentrations to be the common charac-
teristic. The first step toward forming
coherent groups of stations was to do a
histogram-type analysis on the observed
concentrations from the monitors in the
surface network over the model simu-
lation period. Only daytime (0800-1900 h,
LST) hourly values were included in the
histogram analysis. The values used in
the histogram were normalized to
represent the fraction of observations at
each monitoring station that fell within a
given concentration range for the daylight
hours over the model simulation period.
Similar data were compiled for the 214
monitoring sites and were then subjected
to a cluster analysis to form coherent
groups of sites based on frequencies o
observed O3 concentrations. Six suet
groups were formed.
Air quality monitoring for NOX and N0;
is generally not as extensive as that fo
O3 in the U.S. As with 03, nearly all o
the sites were located within urban areas
Using this network to verify the per
formance of the ROM for predicting NO
and NO2 concentrations poses severa
problems. The greatest problem is tha
the nature of these chemical species ir
the atmosphere is largely primary; that is
the flux of emissions of the species i;
generally the largest contributor to the
ambient concentrations. This being the
case, the large spatial variations ir
emissions patterns are expected t<
produce large spatial variations ii
ambient concentrations, especially on the
urban scale where the emissions hetero
geneity is greatest. Urban-oriented moni
tors, such as those available here for NO
and NO2, will capture concentratior
patterns characteristic of the local are;
only. Monitors located in rural areas, o
areas that might be more regionally
representative, were generally not avail
able in 1980. Similar scale problems exis
for 03 but are mitigated somewhat by the
secondary nature of the pollutant and it;
smaller spatial concentration gradients
For the quasi-deterministic analysis usec
for the evaluation, data from NOX anc
NO2 monitors were aggregated ir
individual urban areas with large NO
source emissions. These were majo
metropolitan areas that contained multi
pie monitoring stations. The number o
monitors in each group varied from 2 fo
the Boston and Washington areas to 1(
for the New York area.
Unlike the evaluation analyses fo
surface concentrations of O3, NO2, anc
NOX> the ROM evaluation for surface
NMHC and all aircraft observations ii
more diagnostic than operational. There
are a number of reasons for this. First
the intermittency of the data poses some
problems in setting up evaluative tests
The NMHC and aircraft data are sparse ir
both space and time. Group aggre
gations, as performed for the othe
species, are more difficult to perform
Therefore, the comparisons of ROM pre
dictions and observations are made in <
deterministic manner for diagnostic pur
poses to show whether there is a clea
systematic bias in the way in which the
ROM predictions of surface NMHC anc
concentrations predicted aloft compare
with observations. Close statistical com
parison of the data set is not warranted.
-------
"ummary of Results and
Conclusions
The analysis using standard surface-
monitored concentrations has shown that
ROM2.0 predicts hourly O3 concentra-
tions above 80 ppb fairly well, and
concentrations between 60 and 100 ppb
particularly well. The average percentage
of daylight hours (08-19 h, LST) over the
simulation period showing concentrations
above 80 ppb was 21.7% in the observed
data set and 19.3% in the predicted data
set for station groupings 1-3. These were
the station groups which showed high to
moderate values of observed 03 concen-
trations. This statistic was compiled from
data over the nearly 50 day simulation
period spanning the summer of 1980.
Cumulative concentration frequency dis-
tributions and histogram analyses of
daytime hourly 03 data have shown the
model to typically predict a narrower
overall concentration range than ambient
observational data indicate. High hourly
values are generally underpredicted and
low values are overpredicted. This feature
was evident in nearly all groups of
monitoring stations. The cumulative fre-
quency distributions treat the observed
and predicted data sets independently for
the simulation period. Table 1 compares
ie predicted and observed frequency
jistributions for daylight hourly 03 con-
centrations for all station groups.
The model has shown good perform-
ance in predicting maximum daily 03
concentrations averaged within station
groups during the simulation period. The
average daily maximum 03 concentration
over the simulation period for station
group 1 was 88.1 ppb for the observed
data set and 82.6 ppb for the predicted, a
6.2% underprediction. An examination of
the time series of observed and predicted
maximum 03 concentrations, however,
revealed that for this group there were
often more significant differences
between predictions and observations on
any given day. At the 75th percentile
level of daily maximum 03 concentrations
the ROM tended to underpredict by 30-
50 ppb during episodic periods. For
group 2, the average daily observed
maximum was 76.9 ppb and the
corresponding predicted value was 79.5
ppb, a 3.4% overprediction. For group 3,
the observed maximum was 64.5 ppb
and the predicted maximum was 70.1
ppb, an 8.7% overprediction. The time
series analysis for group 3 showed that,
on a daily basis, the median and 75th
percentile levels of the maximum 03
concentration showed better agreement
between predictions and observations
than did the group 1 results, with both
over- and underpredictions occurring
throughout the simulation period. The
average performance over these three
groups was a 2% overprediction of the
daily maximum 03 concentration.
A key indicator of model performance
on the regional scale is the accuracy of
simulating the spatial extent and location,
as well as the magnitude of the pollutant
concentrations within plumes from sig-
nificant source areas on the regional
scale, such as those emanating from
major metropolitan areas within the
model domain. ROM2.0 performance
analyses in plume simulation were
conducted for the Northeast Corridor sub-
domain of NEROS, including the major
metropolitan areas from New York
through Boston. During episodic condi-
tions during the simulation period, the
urban plumes from Washington, DC,
Baltimore, New York and Boston could be
clearly discerned in the model predic-
tions. There was a systematic under-
prediction in the 03 concentrations
downwind of Philadelphia, often making it
difficult to discern an integral urban
plume. Figure 2 presents an example of
the comparison of observed and pre-
dicted contours of maximum hourly 03
concentrations over an episode.
Aggregate groups of monitoring sta-
tions for NOX and NO2 were formed for
the major urban areas within the model
domain. Results of model performance
analyses for NOX and N02 showed that
the ROM2.0 significantly underpredicts
NOX and N02 among urban station
groups at the 90th percentile of the
cumulative concentration frequency dis-
tribution. Table 2 presents the average
ratio of observed to predicted NOX and
N02 concentrations at the 50th and 90th
percentile levels, averaged over all
station groups. Spatial patterns of 06-09
h, LOT concentrations of NOX and NO2
were also analyzed for the Northeast
Corridor sub-domain. Results of this anal-
ysis demonstrate that concentrations in
this morning period show factors of 2-3
model underprediction. Exceptions occur
in the Toronto and Philadelphia areas
where the model predictions of NOX and
N02 appear to be significantly higher
than than those in other areas. This may
infer overestimates of NOX emissions for
these areas. A confounding factor in the
performance analysis for NOX and NO2
was the scarcity of significant areas of
concentrations above 5 ppb, either in the
model predictions or in the observations,
for large areas of the domain. The
accuracy of standard NOX and NO2 moni-
tors degrades greatly below this level,
and observed data cannot be trusted to
be accurate when concentrations are less
than 5 ppb.
Model performance on surface-derived
NMHC concentrations was performed in
a diagnostic manner because of the inter-
Table 1. Frequency Distribution of Daylight Observed and Predicted Oj Concentrations
Within Receptor Groups for the Period 14 July - 31 August 1980
Percent of Daylight (08-19 h, LST) O3 Concentrations Between:
Group
1
2
3
4
5
6
Number of
Sites
35
39
64
54
20
2
5-20 ppb
Obs.
8
15
16
22
42
0
Pred.
1
1
1
1
2
0
2 1-40 ppb
Obs.
19
23
28
37
38
26
Pred.
7
9
9
12
16
4
4 1 -80 ppb
Obs.
39
41
46
36
19
63
Pred.
66
69
79
80
73
93
> 80 ppb
Obs.
34
21
10
5
1
11
Pred.
26
21
11
7
9
3
Tattle 2. Average Ratio (Obs./Pred.)
Over Station Groups at 50th
and 90th Percentiles of Cum-
ulative Frequency Distribu-
tions
Daytime
08- r 9/7, LST All Hours
Percentile Percentile
NOX
NO2
50th
1.8
2.2
90"1 SQth
2.3 1.9
2.2 2.1
grjth
2.5
1.9
-------
mittency of the measurements. All ob-
served concentrations were calculated
from a sum of the concentrations of
individual species determined from gas
chromatographic analyses of the NMHC
samples. The ratios of observed to
predicted NMHC concentrations aver-
N43.
N42.
N41.
N40.
aged over the 06-09 h, LDT period were
generally in the range of 4-7 for large
urban areas, and the ratio for afternoon
hours outside of major urban areas was
in the range of 1-3. Despite the large
model underpredictions, an analysis of
the carbon distribution among reactivity
N39.
(a)
(b)
Figure 2.
A/42,
N41.
N40.
N39.
Contours of maximum hourly 03 concentrations over the period July 20-22, 1980
for (a) observed and (b) predicted data sets. Contours of observed data are in
concentration units of ppb and contours of predicted data are in units of
ppb/100.
classes in the observed and predictec
data sets showed quite good agreemen
in the relative distribution of carbor
mass between higher and lesser reactive
classes. The biogenic portion of the
NMHCs (as judged by the explicitly
modeled isoprene compound) did no
show up consistently in the measurec
NMHC samples. It was therefore difficul
to judge ROM2.0 performance on ar
absolute basis. The model predictions o
isoprene concentrations, however, were
in good agreement with surface value;
quoted in the literature from special fielc
studies.
Model performance analyses usiru
aircraft data were also performed in ;
diagnostic manner because of the inter
mittency of the flights both in time an<
space. The analysis for 03 concentra
tions shows that, in general, the overal
background values in the planetary
boundary layer do not build up as higl
under episodic conditions in the pre
dictions as they do in the observations
Typically, regionwide 03 concentration;
predicted by ROM2.0 were in the 40-6(
ppb range, while corresponding concen
trations measured by aircraft monitor!
were 40-90 ppb, with the higher ob
served values occurring under strong
episodic conditions. Most instances d
regionally elevated O3 values wen
underestimated by 20-30 ppb. Ozon<
aloft in the morning, upwind of urbai
areas, was generally underpredicted oi
days when aircraft measurement!
showed values greater than -70 ppb
The regional background 03 underpre
diction by the ROM was, however, not i
totally pervasive phenomenon. Exam
ination of predicted spatial pattern:
reveals the model to predict large area:
of >80 ppb concentrations which buil<
up, especially during daytime hours, am
are transported through the mode
domain. These areas often merge witl
each other as they grow larger and an
moved by the transport fields. Typically
they shrink in size after the sun sets am
gradually dissipate over the nighttimi
hours. Ambient patterns, as judged b
the intermittent aircraft observations
appear to show more widespread area
of >80 ppb 03 concentrations unde
episodic conditions. These areas als
apparently do not shrink in size an*
magnitude over the nighttime period a
much as the ROM predictions sugges
as judged by the early morning aircra
flights. Background values of NOX wer
also underpredicted, by factors of 2-J
Observed values of NOX concentratio
were often in the 5-10 ppb range o
-------
regional flights with predictions along the
ame flight path typically less than 5 ppb.
Flights conducted in and downwind of
urban areas showed ROM's 03 concen-
tration predictions to be very credible,
especially when model wind fields
provided transport along the Northeast
Corridor from a southwesterly direction.
The urban plumes simulated downwind of
Boston, New York, and Washington
especially showed good agreement with
aircraft observations. The ROM tended to
underpredict the maximum concentration
areas within the plumes, although the
degree of underprediction was not great
for concentrations as high as 150-180
ppb. For observed concentrations over
200 ppb, as measured by aircraft
monitors, the model underprediction was
generally more substantial. Model per-
formance or NOX concentrations ob-
served in the aircraft data shows ROM2.0
to generally underpredict, with observed
to predicted ratios of + 5 ppb NOX in the
range of 1-2. These ratios were generally
somewhat less than those for the
corresponding NOX ratios in the aircraft
background measurements and for the
surface concentration analyses. Concen-
trations of NMHC aloft during the
morning period were observed to be
considerably lower (by up to a factor of
10 or more) than the morning surface
NMHC concentrations. ROM2.0 still gen-
erally underpredicted the NMHC concen-
trations observed from aircraft samples,
but by a smaller factor (1-3, as opposed
to 4-7 in the surface data).
It is clear from our study that a critical
factor in the utility of regional model
results is the correct simulation of the
location of regional concentration pat-
terns, especially from urban area plumes.
The important factor in the vicinity of
major urban areas is the correct simu-
lation of the mixing of upwind, over, and
downwind air concentrations. Providing
accurate wind fields to regional models
will help ensure this. We are attempting
to correct some transport biases dis-
covered in ROM2.0 from this perform-
ance evaluation. The problem of localized
circulation features that often exist on
scales smaller than the existing meteoro-
logical monitoring network will still
remain. For Northeast Corridor cities,
these local circulations may greatly influ-
ence plume transport patterns. Ulti-
mately, a nested scale predictive meteor-
ological model may be needed to
generate the transport fields used in
ROM to most accurately model plume
transport on regional scales in areas of
complex flow.
While the overall performance of
ROM2.0 for predicting 03 concentrations
has not shown sufficient cause for
rejecting use of the rrtodel in regulatory
analyses,2 a number of issues have
arisen that prevent recommending use of
the model in a simple, unassisted manner
for studies involving violations of the 03
air quality standard. These include day-
to-day variability of model bias results,
systematic transport biases seen in the
spatial patterns, and perhaps most
importantly, a general underprediction of
regional background 03 concentrations
under episodic conditions. For applica-
tions studies, such as the upcoming
ROMNET (Regional Oxidant Modeling for
Northeast Transport) project, the ROM
results must be used with care. Episodes
must be chosen for which the model is
operating at its most consistent level with
best fidelity. Careful analysis of the simu-
lated transport patterns and background
O3 levels must be made for a given
episode before model results can be
used for subsequent analysis. Sensible
guidance for use of ROM results will re-
quire strong interaction of the modelers,
with their insights and ability to interpret
model results, and the regulatory users of
the information.
This model performance study has also
raised some key questions concerning
the data base with perhaps the greatest
inherent uncertainty, the emissions inven-
tory. The large and consistent underpre-
diction of NMHCs seems to indicate de-
ficiencies in the hydrocarbon portion of
the inventory. These deficiencies may be
across the board on all sources since the
2Model evaluation studies, such as this one, will not
indicate model acceptance, but rather will only
indicate whether model rejection is warranted.
Model acceptance comes only after non-rejection
in many evaluation studies ovier tme.
normalized carbon fractions seem to
agree fairly well in the observations and
predictions of NMHC reactive classes.
Missing or incorrect temperature sensitiv-
ities of both biogenic and anthropogenic
precursor emissions may be responsible,
in part, for the failure of the model to
build the background tropospheric 03
concentration levels as high as they have
been observed from field study results
during episodic conditions. Current
efforts toward the development of a 1985
base year emissions inventory by NAPAP
will take steps to rectify some of the
known deficiencies in the 1980 inventory.
To complete the picture of model
performance evaluation, there must be a
corresponding model sensitivity study. A
systematic sensitivity study of a complex
regional model such as ROM is a very
resource-intensive task, and was outside
the scope of the present effort. Results
from a sensitivity study will, among other
purposes, help guide the application of
resources to narrow the uncertainty in
those required data for which the mode!
is most sensitive. Results from a model
evaluation study, such as this one, will
help to provide invaluable guidance
toward establishing some of the critical
sensitivity tests. We hope to begin such a
sensitivity study soon. The combination
of performance evaluation and sensitivity
results is key to understanding why a
model has performed in the manner it
has under varying conditions.
Model performance is a continuing
task. Once a model is used in production
for multiple applications some type of
performance study is required for each
application. Even though the science em-
bodied in the model may be up to
contemporary standards, the stochastic
nature of the atmosphere can cause
variations in model performance from
location to location and from time to time.
Detailed data bases, including aircraft
based measurements, such as those
used in this study, are not commonly
available. However, the methods of eval-
uation demonstrated here may also be
applied to routinely available data bases
for less rigorous evaluation exercises.
-------
The EPA author. Kenneth L Schere (also the EPA Project Officer, see below), is
on assignment to the Atmospheric Research and Exposure Assessment
Laboratory, Research Triangle Park, NC 27711, from the National Oceanic and
Atmospheric Administration, Richard A. Way/and is with Computer Sciences
Corporation, Research Triangle Park, NC 27709.
The complete report, entitled "EPA Regional Oxidant Model (ROM2.0): Evaluation
on 1980 NEROS Data Bases," (Order No. PS 89-200 8281 AS; Cost: $36.95,
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/S3-89/057
CHICAGO
------- |