-------
I
«•*
u
3
I'
2 <„
9)
nils
llBUj?
o>£
c •»•
>
»-
sf s 2
ui 2jro
0)
*J
CO
a
(T3
•a
s-
3
•a
(D
s_
+j
u
13
S-
4->
I/)
s_
01
>»
ITS
03
(J
O)
CM
a;
CD
-------
lations) that the ROM wind field processor can generate from a given set of
hourly wind observations. These multiple wind fields are each consistent
with both the observations and physical laws governing atmospheric flow and
are assigned probabilities of occurrence based on the inherent kinetic
energy contained within the field. From the resulting set of wind field
realizations used in a corresponding set of ROM simulations the model gener-
ates a concentration probability distribution for a given receptor site
within the domain instead of a single concentration value. When all members
in the set of realizations are considered concurrently, this aspect of the
model prediction makes the ROM different than most other air quality simula-
tion models and presents more of a challenge to the evaluation effort.
The first and second moments of the predicted concentration distribu-
tion can be used to determine the expected frequency with which the observed
concentrations should fall within a given interval. The utility of the
model for regulatory use is measured partly by the width of this interval.
The second moment of the distribution (the concentration variance) is the
parameter that defines this width. It is a measure of the inherent trans-
port uncertainty of the model. If the width is very large, the model may
provide no more information than one would gather by guessing the expected
concentration. This is true even if the model is shown to be accurate in
other respects. Therefore part of the ROM evaluation should consist of an
analysis of the predicted concentration variances. For regulatory use, the
predicted concentration for maximum ozone at a given receptor location for
a single realization is the primary parameter of interest. This value may
be compared to the corresponding observed value. Also, if groups of recep-
tors can be identified having similar predicted concentration distributions,
the observed distribution composed of the measurements from each receptor
within the group can be compared to the predicted distribution from the
group. Such groups might be stratified by receptor location with respect
to downwind distance from major source emissions areas. Section 4 dis-
cusses these and other potential evaluation methods for analyzing the ROM
predictions in the NECRMP application.
-------
Base case evaluation studies with the ROM will be conducted for 4 or 5
regional smog episodes from the data base. The individual episodes range
from 2 to 6 days in duration, and are mostly from the 1980 summer period.
The exact choice of modeling episodes is tentative as of yet because the
analysis of the air quality data base has only recently begun. A full 2
week simulation will also be performed.
-------
SECTION 2 -
THE ROM SYSTEM
OVERVIEW
The ROM described here is the 2nd generation version of the model.
It is designed to simulate air pollution chemistry and dispersion over a
1000-kilometer-by-1000-kilometer area for multiple day periods. The
simulation of photochemical air pollution over such scales is a complex
problem. The ROM reflects this complexity, consisting of some 25 programs
accessing almost 100 files. The overall structure of the data flow for the
ROM modeling system is illustrated in Figure 3.
The input files contain the "raw" data accessed by the modeling system.
Actually, these data files are produced by a combination of manual and
computer manipulation of raw data. However, the input data files consti-
tute a boundary between the ROM and the rest of the world, and therefore
define the input data for the model. The initial preprocessors take the
"raw" input data and transform it for use by other preprocessors. These
data are transferred between preprocessor programs through the processor
input files (PIF). The preprocessor programs are used to develop intermedi-
ate parameters that are then transformed into core model inputs. Examples
of these intermediate parameters are the wind fields, emissions, and turbu-
lent fluxes. The preprocessors transfer the processed data through the PIF
files and the model input files (MIF).
The data contained in the MIF files are converted into the form re-
quired by the CORE model through the execution of the b-matrix compiler
(BMC) program. The BMC translates the parameter fields in the MIF files
(layer thicknesses, horizontal winds, interfacial fluxes, deposition veloci-
ties, etc.) into the matrix and vector elements necessary to operate the
CORE model. The BMC transfers the information needed by the CORE model via
the core input files.
-------
c
AIR QUALITY
METEOROLOGY)
EMISSIONS
CLOUD COVER
c
LAND USE
TOPOGRAPHY
PROCESSOR
NETWORK
PROCESSOR
AND
MODEL
INPUT
FILES
V INITIAL \
AND \
BOUNDARY /
:ONDITIONS/
ALGORITHMS
FOR SOLVING
GOVERNING
EQUATIONS
^/CONCENTRATIO
\ OUTPUT
CHEMICAL
KINETICS
ALGORITHM
Figure 3. Data flow for the ROM system.
-------
The CORE model is the computer language- analogue of the differential
equations that describe the processes involved in the chemistry and disper-
sion of regional photochemical air pollution. The CORE model is expressed
in fundamental mathematical form. All inputs to the model are matrices and
vectors whose elements are composites of meteorological parameters, chemical
rate constants, etc.
The output of the CORE model consists of the layer-averaged pollutant
concentrations for every grid cell every 30 minutes. However, these data
are not the final model results. The ROM incorporates two new concepts in
air pollution modeling. The first is a method for simulating the physical
and chemical processes that occur within about the first 100 meters of the
ground. The result of this scheme is that both the ground-level concentra-
tion and the root-mean-square concentration variation within each selected
cell are produced. These data can be extracted using the results of the
CORE model output (contained in the model output files) and parameters
contained in the model input files for a time period of 30 minutes or
longer.
The other concept is the incorporation of the uncertainty of the wind
field representation. It is known that many different wind fields can be
constructed that match the observed wind data and empirical and theoretical
constraints. In the ROM, a family of possible wind fields is used in the
model and leads to a result of an ensemble of concentration fields. There-
fore the model results will consist of a distribution of concentration
values and associated probabilities of occurrence for each grid cell for
each time step. The implementation of the multiple wind field concept
requires that the model be run many (~10) times for each simulation.
ROM postprocessor programs will be used to transform the results of the
individual CORE outputs into ensemble concentration distributions. Other
model postprocessors will transform the layer 1 concentration values into
ground level (layer 0) concentration distributions. These data constitute
the model results. Measures of model accuracy will be made by comparing the
model predictions of concentration probability distributions with observed
-------
measurements as well as examining individual model runs (realizations) on a
single wind field.
PREPROCESSOR DESCRIPTIONS
The model preprocessors transform the "raw" input data into the matrix
and vector elements required by the CORE model. Some of the preprocessors
are quite simple; others such as the wind field processors are complex
computer models in themselves. All of the preprocessors communicate with
each other along data flow channels through the use of the PIF and MIF
files. The preprocessor system is designed to proceed in stages (0-6) with
data developed in one stage used as input for the next stage. Figure 4 is
a schematic diagram of the network of ROM processors and data channels. In
this section an overview of the model processors proceeding from stage 0
through stage 7 is presented.
Stage - 0
Processor P03 - This processor performs standard operations on the
surface meteorological data to reformat them for the higher level processors
in the network. The processor reads the surface meteorological data that
are usually available every three hours and interpolates the data into
hourly values. Some stations report every hour, in which case temporal
interpolation is not necessary. The variables produced by this preprocessor
include surface wind velocity components, surface mixing ratio, relative
humidity and dew point, surface virtual temperature, geopotential of the
1000mb surface, time rate of change of the 1000mb geopotential, sea level
pressure, surface potential temperature, and the surface air density.
Processor P25 - This processor converts land use percents into
land use fraction for each grid cell in the model region.
Processor P05 - This processor transforms the raw cumulus cloud
digitized satellite data into hourly gridded sky coverage of this type of
cloud.
10
-------
(U
CO
-M
C
Ol
c
o
Q.
O
O
O
c
o -*
•£> O
ro S
4-> OJ
!— S-
i— O
•I— CO
to
o oj
•r- U
o
Q.
03
O O
oo o:
CLl
S_
3
C7)
11
-------
Processor P06 - This processor reads the input terrain elevation data
and computes the average elevation for each cell in the modeling region.
The preprocessor also computes smoothed elevations for use by the P08
processor.
•
Processor P13 - This processor computes the gridded line source length
totals for each cell.
Processor P14 - This processor prepares the files containing stack
descriptions and emissions for all major point sources, and hourly gridded
emissions for all other sources.
Processor P16 - This processor interpolates the rawind upper air data
into hourly observations.
Processor P19 - This processor reads the hourly surface meteorological
observations and computes hourly gridded fraction of cloud coverage for all
cloud types combined.
Processor P20 - This processor uses the ROM grid definition to produce
required matrix elements for Pll when it tranforms gridded divergence data
to Fourier series expansions.
Processor P21 - This processor contains indices that classify each grid
cell in each layer to one of 5 possible sets of initial concentration profiles
based on the magnitudes of measured pollutant loadings of 03 and precursor
species.
Stage - 1
Processor P01 - This processor performs a number of standard operations
on the interpolated hourly rawind data for use by other, higher level, pro-
cessors. Wind vectors, mixing ratio, relative humidity, dew point, temperature,
potential temperature, static stability, air density, and pressure are computed
12
-------
for every 50 meters from the surface to 5000-meters. The geopotential of the
850, 700, 500, and 1000mb surfaces are al-so computed.
Processor P02 - This processor determines the initial concentrations
throughout the grid and the upper and lateral concentrations at the grid
boundaries for all pollutant species simulated by the CORE model. The first-
generation model assumes typical "clean atmosphere" concentrations for species
entering the grid boundaries. The initial concentrations are specified using
five sets of preequilibrated initial concentrations. Clean atmosphere concen-
trations are used to define rural initial conditions. For grid cells in or
near urban areas, observed data are used to guide the manual selection of one
of the five sets of initial concentrations that best represents the initial
conditions in that grid cell. The second-generation model will likely employ
the same procedure. The matrix defined in P21 classifies each cell according
to 1 of the 5 sets of initial concentrations.
Processor P04 - This processor computes the surface roughness, Obukhov
length, surface heat flux, and the friction velocity for each grid cell for
each hour.
Stage - 2
Processor P15 - This processor computes the effective deposition resis-
tance and deposition velocity for each pollutant species every hour for each
grid cell. The processor uses the surface roughess, friction velocity and
Obukhov length computed by processor P04 in stage 1.
Processor P07 - This processor computes terrain penetration factors
(Lamb, 1983), the height of the surface based inversion layer, and the wind
field in the surface inversion. Within the processor is a complex numerical
model that simulates the motion of a viscous, hydrostatic fluid of constant
density over irregular terrain.
13
-------
The processor first calculates Ap^. This parameter is used as an
indicator of the presence of a surface ba-sed inversion (1, if surface based
inversion is present, and 0, otherwise). The value of API is computed
hourly for use by the wind field processor Pll. The model assumes that an
inversion layer forms over the entire model domain at the hour that the
surface heat flux averaged over the whole region becomes negative, and it
disappears everywhere at the hour that the mean "temperature deficit" (de-
fined by equation 7-11, Lamb, 1983) first becomes negative.
For time steps occurring when API equals 1, preprocessor P07 com-
putes the growth of the surface inversion and the layer averaged velocity
field within the inversion. Since the surface wind observations during
these conditions either are erratic or reflect purely local conditions, the
inversion layer flow model uses observed pressure gradients to initialize
the flow field and then drives the flow model using observed temperatures.
No wind observations are used in the calculation of the inversion layer
wind field.
For all time steps, processor P07 computes the elevation of the top of
layer 1, and sets the depth of layer 0 to one-tenth of the value of layer
1. The processor also converts the height of layer 1 and the height of
the virtual surface into pressure coordinates. The height of the virtual
surface is the terrain elevation plus the depth of the surface inversion
when an inversion is present, and is equal to the terrain elevation other-
wise.
A final computation performed by this processor is the estimate of the
fractions of the top surfaces of layers 0 and 1 that are penetrated by
terrain. Currently the value of the terrain penetration fraction for layer
0 is set to the value calculated for level 1.
Processor P18 - This processor computes matrix elements needed for Fourier
series expansions of wind fields evaluated at all rawin and surface meteor-
ology stations.
14
-------
Stage - 3
Processor P08 - This processor determines the elevations of the top
of layers 2 and 3, computes the mean vertical velocities on these surfaces,
and estimates cloud updraft velocities and other parameters (listed below)
needed to specify pollutant fluxes across the surfaces of layers 2 and 3.
The following parameters will be calculated by processor P08 every
time step for each grid cell in the model domain: the plume entrainment
velocity, the cumulus updraft speed, the vertical air speed on the inter-
face between layers 2 and 3, the elevation of the interface of layers 2 and
3, the elevation of the top of the grid (the top of layer 3), the volume
flux through the top of the grid, the local time derivative of the elevation
of the interface between layers 2 and 3, the thickness of layer 3, the
thickness of layer 2, the pressure at the top of layer 3, the pressure at
the top of layer 2, and an estimate of the cumulus flux partition function
(see equation 5-44, Lamb, 1983).
Stage - 4
Processor P09 - This processor prepares the information necessary to
correct the chemical reaction rate constants for variations in atmospheric
density, temperature, cloud cover, and solar zenith angle. Layer-averaged
density correction factors and layer-averaged virtual temperatures are used
by the CHEM portion of the CORE model to correct the chemical reaction rate
constants. The solar zenith angle and the cloud cover correction factor
are used to modify the photolytic rate constants used in CHEM.
Processor Pll - This processor generates a family of vertically inte-
grated horizontal winds for each of the model's layers. During the daytime
hours (i.e., when API = 0 this processor produces wind fields for each of
the model's three layers. At night (i.e., API = 1) the processor produces
wind fields for layers 2 and 3 only. The flow field generated by the P07
processor for layer 1 is passed to Pll for combination with the wind fields
derived there.
15
-------
This processor uses a novel numerical scheme to produce layer-averaged
wind fields that are: (1) consistent with observed wind data, (2) consistent
with the mass-continuity equation and derived gridded wind field divergences,
and (3) consistent with the theory and observations that, in two dimensional
fluid flow, kinetic energy is partitioned among the spatial fluctuations in
the flow proportional to the inverse cube of the wave number of the fluctua-
tions (minus-3 law).
Pll calculates a family of wind fields consistent with the observations
and theory. The base wind field, W0, is obtained by calculating a wind
field whose average kinetic energy is the same as the average kinetic energy
of the observations, whose components match the measured winds, and whose
divergence is consistent with specified gridded wind field divergences. An
iterative convergence procedure is then used to search for a wind field
that obeys the minus-3 law and is close to the initial wind field solution,
W0. This wind field is designated as field A.
The next four wind fields, B, C, D, and E, are computed by moving a
specified distance in a random direction from the first wind field, A, (in
phase space) and then projecting back to the energy sphere whose intersec-
tion with the solution plane contains the first wind field solution A.
This procedure results in wind fields whose average kinetic energy is
identical to the first wind field solution but whose kinetic energy parti-
tion does not exactly obey the minus-3 law. Wind fields F, G, H, I, and J
are obtained by starting the iterative convergence procedure at locations
that are selected by moving a specified distance on the solution plane (in
phase space) in a random direction from the initial solution point, W0.
The random displacement is the same as for solutions B, C, n and E and was
determined by experimentation. Wind fields F, G, H, I, and J have average
kinetic energies that differ from observations but the kinetic energy is
partitioned in a manner consistent with the minus-3 law. A detailed descrip-
tion of the methodology in Pll can be found in Lamb (1984). A graphical
illustration of the wind field calculation procedures are presented in
Figure 5.
16
-------
SOLUTION PLANE
ORIGIN
(KINETIC ENERGY = OBSERVED)
APPLY POWER LAW CONVERGENCE
TO GET SOLUTION A.
SOLUTION PLANE
• ORIGIN
SOLUTIONS B,C,D, AND E:
GO OUT DISTANCE § IN A RANDOM DIRECTION
FROM A; THEN PROJECT BACK TO 'ENERGY
SPHERE'CONTAINING A.
SOLUTION PLANE
ORIGIN
SOLUTIONS F,G,H, I, AND J:
GO OUT DISTANCE 5 IN A RANDOM DIRECTION
FROM W0, THEN APPLY POWER LAW
CONVERGENCE TO GET F TO J.
Figure 5. Illustration of the wind
solution schemes.
field processor (Pll)
17
-------
The ROM will be run ten times, with wind fields A through J, for each
scenario simulation. The results of the individual runs and their associ-
ated probabilites will be used to construct the concentration probability
function at each time step and grid cell.
During daylight hours the surface wind velocities are used as input to
the calculation of the layer 1 averaged wind fields. The upper air wind
velocities are used as input to the calculation of the layer 3 averaged
wind fields. Both surface and upper air observations are used for layer 2.
Because of the numerical method employed in this processor no pair of
surface wind observations can be closer than about 25 kilometers. Since
there are a large number of surface meteorological stations in the modeling
region some surface stations will be excluded based on the 25 km criteria.
In practice, this means that the availability of surface wind observations
is not likely to be an important criteria in the selection of model valida-
tion scenarios. All upper air data within the model will be used for the
wind field calculations.
Processor P12 - This processor calculates the parameters required to
describe the material flux between layer 0 and layer 1 and between layer 1
and layer 2. It also calculates the horizontal eddy diffusivities in all
layers.
Processor P17 - This processor calculates the local time derivatives
of the cell heights in layers 1 and 2.
Stage - 5
Processor P10 - This processor transforms the emission data into the
source strength functions required by the b-matrix compiler. The source
strength functions partition the point source and area emissions into
layers 0, 1 and 2. Processor P10 also computes the plume volume fraction
which is used in the layer 0 equations to parameterize subgrid scale chemi-
stry effects.
18
-------
Processor P29 - This processor transforms the velocity and diffusivity
fields into parameters required by the CORE model. The velocity, diffusivity
and cell volumes are used to compute an effective velocity field that is
then used to compute "back track" trajectories. The coordinates of the
"back track" trajectory origin associated with a given cell for a given
time step are used by the CORE model to solve the advection-diffusion
equation.
Stage - 6
Processor P28 - This processor, the b-matrix compiler (BMC), trans-
lates the variables residing on the MIF files into the elements of the
"b-matrix" defined in Chapter 9 (Lamb, 1983). The variables prepared by
processor P29 are passed through the BMC and saved on the file MFBMAT.
The MFBMAT file and the MFIBC file written by the initial and boundary
condition processor (P02) constitute the input data for the CORE model.
CORE MODEL (Stage - 7) AND POSTPROCESSORS
The CORE model is the main program of the ROM. It reads the MFIBC and
MFBAT data files and computes the concentration of each pollutant specie in
layers 1, 2 and 3 in each grid cell for each time step (30 minutes). Since
the CORE model operates on one wind field realization at a time the model
will be run many times (~10) for each simulation scenario.
The concentration distribution functions, taking into consideration
results from all realizations, for selected pollutant species in layers 1,
2 and 3 in each grid cell at each time step will be calculated by a postpro-
cessor. This program will read the concentration fields produced by the
individual CORE model runs and the probability weighting functions assigned
to the various realizations (Lamb, 1984). Pollutant concentrations in each
cell for each time step are described by a concentration value from any
given realization or a concentration distribution when values from all
realizations are considered together. Concentrations in layer 0 are computed
19
-------
using the current concentrations for layer l.and parameters contained in
the MIF files. The equations used for the evaluation of the layer 0 concen-
trations and their root-mean-square variation are given in Chapter 5 (Lamb,
1983).
•
As an example of how input data flows through the preprocessor chain,
a section of a data stream, or channel, in the ROM system is given in Table
1. The section that appears there follows the surface and upper air meteoro-
logical processing from the raw input data through the first 2 stages of
processing. Data processing must occur in 6 stages of processors before
the simulation model can be executed. The processed parameters shown in the
table would flow through principal and branch channels for all stages.
Similar data channels mark the flow of air quality, emissions terrain and
land use, and miscellaneous parameters. Lamb (1984) discusses the data
channels and the details of the data processing in detail.
20
-------
TABLE 1. SAMPLE DATA CHANNEL IN ROM SYSTEM
SELECTED STAGE 0 PROCESSORS
Raw Input
SID
CLD
P.Ps
WS
WD
TJd
UID
Pa
Ta>Tda
RHa
wsa
WDa
Description Processor Output Parameter Description
surface station
id and location
surface observations
of cloud cover,
amount, heights
\
P3 u,v wind components
P19
(Surface)
Meteorological
Processors
q mixing ratio
RH relative humidity
Td> Tv> Tp dew point, virtual,
and potential
station pressure and ) } temperature
sea level pressure
surface wind speed
surface wind direction
surface temperature
and dew point
height and rate
(j>o, <{>0 of change of
1000 mb surface
Ps sea-level pressure
p air density
/ / ac fractional coverage
upper air station
id and location
pressure level aloft
height aloft
temperature and
dew point aloft
relative humidity
aloft
wind speed aloft
wind direction aloft
P16
(Upper Air
Meteorological
Processor)
of all cloud types
ua,va wind components
Ta,Tda aloft
Pa
parameters above are
interpolated regularly
in time
SELECTED STAGE 1 PROCESSORS
Input Parameter Processor Output Parameter
u,v
q
RH
' d ' v ' p
Ps°'
Q
P4
(Surface
Meteorological
Processor)
L
zo
u*
0
Description
Obukhov length (gridded values)
surface roughness (gridded values
friction velocity (gridded values)
surface heat flux (gridderi values)
21
-------
TABLE 1. (continued)
SELECTED STAGE 1 PROCESSORS
Input Parameter
ua> va
Ta» Tda
Processor
PI
(Upper Air
Meteorological
Processor)
Output Parameter
Tda> Tva> Tpa
la
RHa
850. *700« $500
Pa
Description
height aloft
wind components aloft
dew point, virtual, and
potential temperature
aloft
mixing ratio aloft
relative humidity aloft
pressure level aloft
height of 850, 700,
500 mb surfaces
air density aloft
static stability aloft
parameters above interpolated
regularly in vertical space
and time
22
-------
SECTION 3 '
THE NECRMP/NEROS DATA BASE
The NECRMP and related field studies (NEROS I, NEROS II, and PEPE)
conducted during the summer periods of 1979 and 1980 provided a wealth of
meteorological and air quality data over the Northeast U.S. These data
include measurements made specifically for the field studies as well as
routine monitoring of the atmosphere from existing networks. Measurements
above the surface were obtained from instrumented aircraft, balloon-borne
ascending and tethered instrument platforms, and satellite imagery. Surface
measurements were made from fixed monitoring sites as well as special mobile
vehicles.
Table 2 shows the instrumentation and parameters measured from the
principal aircraft used in the NEROS experiments. During NEROS I (1979) the
flight plans were designed to provide Lagrangian sampling of pollutants
within specific air parcels over the regional (1000 km) domain. Measure-
ments were made at several altitudes within the daytime boundary layer and
aloft above the surface-based inversion at night. Aircraft flights during
NEROS II and the NECRMP urban field studies (1980) concentrated on sampling
in and near the urban plumes of Columbus, OH, Baltimore, MD, Washington,
DC, New York, NY, and Boston, MA (Possiel and Freas, 1982) as well as
Lagrangian regional scale sampling. Figure 6 shows a typical Lagrangian
sampling flight track. Air parcel movement was monitored with tetroons.
Over 200 separate flights were made during the 2 NEROS projects over the
July-August periods of 1979 and 1980. Although it is not indicated in
Table 2 hydrocarbon samples were also obtained from the aircraft. These
were analyzed by gas chromatography for individual organic species concen-
trations. An attempt was made to obtain samples over rural as well as
urban areas. The hydrocarbon samples obtained aloft are a supplement to
those obtained from surface sites across the regional domain.
23
-------
TABLE 2. NEROS AIRCRAFT INSTRUMENTATION
NEROS I Aircraft Operator
Research Triangle Institute
Brookhaven National Laboratory
Washington State University
Element1
03
NO/NO*
S02
CCN
8-scat
AT
OPT
03
NO/NO,,
S02
B-scat
AT
OPT
RH
TSR
UVR
03
NO/NO*
S02
CCN
B-scat
AT
RH
'•Oo Ozone
NO/NOX Nitric Oxide/Oxides of Nitrogen
S02 Sulfur Dioxide
B-scat Light Scattering Coefficient
AT Air Temperature
DPT Dew Point Temperature
TSR Total Solar Radiation
UVR Ultraviolet Radiation
CCN Cloud Condensation Nuclei
^Information Not Available
Equipment
Bendix 8002
Monitor Labs 8440
Meloy 285
Environment One Rich 100
MR I 1550 B
Rosemount 102
EG & G 880
AID 560
TECO 14d
Meloy 165
MRI 1550
Yellow Springs 705
EG & G 137-3C
Weather Measure HM 111
Eppley Pyranometer 8-48 A
Eppley UV Radiometer
Bendix 8002
Monitor Labs 8440 HP
TECO 43
Environment One Rich 100
MRI 1550
Metrodata M-8
Metrodata M-8
24
-------
TABLE 2. (continued)
NEROS II Aircraft Operator
Environmental Monitoring Incorporated
AeroVi ronment
Stanford Research Institute
Element*
NO/NOX
S02
S04
B-scat
AT
DPT
03
NO/NOX
S02
S04
B-scat
AT
DBT
NO/NOX
S02
S04
B-scat
AT
DBT
Equipment
Dasibi
Monitor Labs 8440
Meloy 285
Meloy SA285
NA2
NA
NA
Bendix 8002
Monitor Labs 8440
TECO 43
(not measured)
NA
NA
NA
Dasibi 1003 AAS
(not measured)
Meloy SA285E
(not measured)
NA
NA
NA
25
-------
Figure 6. Typical Lagrangian sampling flight track along the East
Coast corridor. Lettered sections indicate cross-
trajectory traverses. Numbers are sampling times.
26
-------
Meteorological data aloft were obtained.from radiosonde releases at
regular National Weather Service (NWS) upper air stations within and outside
the boundaries of the ROM domain. Releases are normally made at 0000 GMT
(1700 EST) and 1200 GMT (0700 EST) on a routine daily basis from these
stations. Additional releases were made at 0600 GMT (0100 EST) and 1800 GMT
(1300 EST) for 7 stations within the ROM domain during the NECRMP study.
Figure 7 shows the locations of these upper air sounding stations that are
used in the ROM data processing. Additional soundings are also available
from the NECRMP urban field studies program (Possiel and Freas, 1982).
Photographs of cloud cover over the model domain made by satellite observa-
tion are another form of non-surface based meteorological measurements used
in the data processing for the model. These provide measures of the total
cumulus cloud cover and the maximum depth of these cloud elements which are
required model input parameters.
Surface based meteorological data were obtained primarily from the more
than 100 NWS stations in the U.S. and southern Canada within and close to
the boundaries of the ROM domain. Figure 8 shows the distribution of the
surface stations providing data for one particular hour. Hourly observa-
•
tions of temperature, dew point, station pressure, wind speed and direction,
and precipitation amount are available. Other observations, some of which
are subjectively determined, include the amount of sky cover, cloud type
and altitude, and prevailing visibility. Most of the meteorological sta-
tions report on a 24-hour basis although a few send reports only during
daylight hours. The large number and high density of surface meteorological
stations within the ROM domain provides more than adequate resolution for
interpolating the required variables over the grid network.
Air quality measurements from surface-based monitoring locations were
derived from three major data source categories. In addition to the routine
measurements available from the SAROAD network described below, two special
monitoring networks were established during the NECRMP study. The special
networks included several monitoring sites for 03, NO/NOX, and hydrocarbons
in the vicinity of the Columbus, OH and Baltimore, MD urban areas as part
27
-------
c
•i—
lO
•M
C
•r—
-C
c
o
(Q
4J
O1
C
-a
c
(O
s-
cu
Q.
a.
o
to
c
o
tO
O
O
CD
S-
28
-------
o
T3
o
a:
-C
4->
's
(O
-M
in
s-
o
V
•r
(O
u
•^
CD •
O S-
r- 3
O O
S- J=
o
0) S-
+-> re
a; •—
v •
O) -r-
O -(->
(O S-
<4- (O
J- a.
O 03
cn CD
C C
O M-
•r- +J
4J S-
(C O
o a.
o OJ
_i s.
oo
01
s_
3
CD
29
-------
of the 1980 NEROS II project to study urban plume transport. Also, during
the 1980 NECRMP urban field studies several monitoring stations were estab-
lished in the vicinity of major Northeast urban areas including Washington,
DC, Baltimore, MD, New York, NY, and Boston, MA. These sites supplemented
•
the existing networks with 03, nitrogen oxides, and hydrocarbon measurements
(Possiel and Freas, 1982). The quality assurance activities conducted for
the surface monitoring programs of the NEROS and NECRMP studies were quite
extensive. For all instruments these activities included frequent systems
audits of monitoring sites, verification of calibration procedures, instru-
ment calibrations and zero/span and precision checks, and instrument perfor-
mance audits. The 1980 field study period will be the focus for obtaining
data for the ROM evaluation because of the greater data availability, the
coincident NECRMP urban field studies on the east coast, and the establish-
ment of a base year emissions inventory for 1980.
The majority of available surface air quality monitoring stations are
part of the established network of the Storage and Retrieval of Aerometric
Data (SAROAD) system. This system assimilates air quality data from the
networks of the National Air Monitoring Stations (NAMS) and the State and
Local Air Monitoring Stations (SLAMS) and includes approximately 5,000
individual sites in the U.S. Hourly average 03 values are available from
the majority of these sites, while nitrogen oxides and hydrocarbon measure-
ments are obtained at fewer locations than 03. Quality assurance procedures
are part of the SAROAD system. This minimizes the amount of erroneous data
entering the system, although the number and frequency of audits on such a
large network of regular monitoring stations are considerably fewer than
those performed on the special NEROS and NECRMP sites specifically set up
for those projects.
Figure 9 shows the distribution of surface 03 monitoring sites included
in the NECRMP field project and Figure 10 shows the regular sites for 03
monitoring in the SAROAD network that are included within the ROM domain.
A complete list of the SAROAD monitoring stations that measured 03 appears
in Appendix A.
30
-------
Figure 9. Locations of monitoring sites for CU operating as
part of the 1980 NECRMP Urban Field Studies.
31
-------
Figure 10. Locations of regular SAROAD monitoring sites for CU
located in the northeast U.S. operating during 1980.
32
-------
The final component of the data base is the source emissions inventory.
Detailed data are required by the ROM to describe hour by hour emissions of
gaseous species from anthropogenic and natural sources within the modeling
domain. The primary species of interest are CO, NOX, and speciated categories
of natural and anthropogenic hydrocarbons. As part of the NECRMP project a
complete emissions inventory for 1979-1980 for all point and area sources was
prepared with the cooperation of the U.S. EPA and all of the states within
the ROM domain (EPA, 1982). These data are being incorporated into the newer
emissions inventory of the interagency National Acid Precipitation Assessment
Program (NAPAP). The base year for these emissions data is 1980. The major
data components in the inventory are U.S. and Canadian anthropogenic sources
and U.S. and Canadian natural (biogenic) sources. The spatial extent of the
NAPAP inventory is sufficient to provide complete coverage of the ROM grid.
An EPA-developed processing routine transforms the annual emissions of the
inventory into the hourly, speciated emissions used by the ROM emissions
preprocessor. Temporal and spatial allocation factors for each source type
and emitted species are used in the transformation (Sellars et al., 1985).
The availability of required input data as well as the availability of
observed air quality data with which to compare model predictions will guide
the selection of specific scenarios for modeling. Further, since the pri-
mary interest in the ROM is in the prediction of photochemically produced
pollutants, the highest priority for modeling scenarios will be for regional
smog episodes as determined from a distribution of observed hourly 03
concentrations. Figure 11 shows a frequency distribution of surface 03
concentrations over the SAROAD and NECRMP monitoring stations within the
ROM domain during the 1980 NECRMP/NEROS study period. A number of high-03
episodes are evident within the period from July 23 through August 16.
This period also lies within the intensive field efforts of the NECRMP and
NEROS studies providing an enhanced data network of meteorological and air
quality observations. Selected episodic periods of 2 days or longer within
the above time "window" will be chosen for model simulation for the ROM
33
-------
1 1 1 1 '
1 1 1 1 1 —
r~ i ' i 1 1 —
- ( i [ i 1
i i I ] 1-
=3 =1 | i . ,
-JU1 ^3 1 '" 1 1 1- <
^ ^ _| ^ ^....., ........ f [ 1 J 1 .
"^ H» d ^ * ' ^*
3«*5 < = . _ i i l i
«= = <13 — i™ 1 — i -i — H
So— = a i • i i
S<5«g tS j= | j | J 1 —
l — I I 1 — 1 1 1 — 1 — 1 1 —
' ' *
\ ' 1 1 1 1 \
i i i i .
i 1 ! e- \
~ i • •'- -• • — "• I l 1 l
- i 1 i | 1 -
1 1 I "I 1
. ,. 1 1 t .
\ ... | 1 1 -1 •'
— 1 1 ' * j —
- 1 1 1 1 1 -
1 III 1 -
1 1 1 J 1
i 1 1 J 1 -
j i i i —
• j i i i
- i \ i i 1 -
i 1 i i
- i . . . [ i v . . . — ^ _,
ocacacoq-cMoaotOtfcMC
CMCM*— T— *— ;«— *-OO^O
00 OOOOOOOOO
m >
en
CM
CM
in
CM
n
CM
O)
ex
un t
<
«—
^
t"
en
IA
1
a,
EM
LO
OJ
^
CM I
CM
cn
^
^
in '
i
i
9
3
3
c
u>
>.
o
M- •
-a
t/l O
C -r-
O i-
•r- 0)
-4-> a.
to
0) -M
O (/)
e
O oo
o o
DC
CTUJ
o z:
3 O
O UJ
M- O
O 00
•r- O)
•»-> J=.
3 4J
.a
•r- CJ)
S- E
+-> -I-
to s-
•I- 3
-a T3
O S-
s= o
cr a)
OJ E
s_
<4- en
c
-t-> -1-
O i-
r- O
C.-M
I -I—
X 5=
o o
CQ E
OJ
S-
O)
34
-------
evaluation study. Also, a longer-term period of approximately 2 weeks from
within the window may also be simulated without reinitializing the model at
any point after the start of simulation.
The emphasis in the model evaluation will be on 03 predictions, al-
though there are other pollutants also of interest. Some study may be
given to the ROM's ability to predict CO concentrations within the domain.
The nearly inert character of CO makes it a quasi-tracer species with which
to evaluate the transport portion of the model without the interference of
non-linear chemistry effects. Also, the primary importance of N02 in the
photochemical cycle suggests that some evaluation statistics may be de-
veloped for this pollutant as well.
35
-------
SECTION 4 .
MODEL EVALUATION
OVERVIEW
The comprehensive evaluation of a 3-D gridded air quality simulation
model is a complex task. At a basic level it is necessary to test whether
the mathematical representations of the individual physical and chemical
processes are correct. This is done for each component process in the ROM.
The validity of the chemical kinetic mechanism, for example, may be tested
with data from controlled smog chamber experiments. The vertical cloud
flux algorithm can be validated with aircraft measurements of material
below, within, and above a convective-type cloud from the NEROS field
program. The wind field algorithm cannot be precisely tested in an inde-
pendent fashion, although individual trajectories can be evaluated from
tetroon data from the NEROS study and the individual wind realizations can
be analyzed for their consistency with measurements from fixed monitoring
sites. Lamb (1983) describes some of these independent tests on component
parts of the ROM during the developmental phase. The evaluation outlined
here considers the model as a whole and attempts to compare its predictions
with ambient observations.
This view would suggest that the ROM could then be viewed as a "black
box" model producing results that could be statistically analyzed against
observations. Although it is tempting to take this view, it must be
avoided. The blind application of statistical tests will not provide
sufficient understanding of model performance to draw any meaningful con-
clusions from the study. The goal of a model evaluation study, such as the
one proposed here, is to gain insight into the model predictions and the
observed data. One wants to know if the model is producing the right
answer for the right reason and whether the model prediction is good enough
for the user's purposes. Dennis and Irwin (1985) argue that model evalua-
tion is as much art as it is science. They advocate that the careful
36
-------
analysis of observational data will provide the direction for a meaningful
path through the course of a model evaluation. The proper choice of statis-
tical comparisons between observations and predictions can be established
from an understanding of the phenomena shown by the data sets. This is the
approach that we take in the ROM evaluation.
The approach will consist of 3 levels of analysis. The first level
will explore the observed data set with diagnostic tests to bring out the
important features, both on temporal and spatial scales, and to order the
data according to the features that are found. Because the principal
pollutant of interest included in the ROM simulations is 03 the analysis of
observed data will focus on it. The second level of analysis will be an
evaluation of the ROM results for an individual realization of an episode.
This approach is a deterministic analysis of each simulated realization
without regard to the other members in the family of realizations. Finally,
the third analysis level will consider concurrently the entire family of
simulated realizations and the resulting probability distributions of
concentration. This analysis structure is actually a hierarchy with the
results obtained at each lower level guiding the steps taken at the next
•
higher level.
DATA BASE ANALYSIS
A thorough analysis of the existing ambient data base is a key step in
any model evaluation study. This analysis should elicit the major phenome-
nological features on the spatial and temporal scales of the model domain.
In turn, the results obtained in this step will provide guidance in the
model evaluation steps. The major thrust in the data base analysis will be
on air quality data from surface monitoring sites. Pollutants of particular
interest include 03, CO, NO, N02, and hydrocarbon species. Air quality
data obtained aloft will provide some idea of the model's predictive abili-
ties in the upper mixed layer and cloud layer, but these measurements were
obtained at irregular intervals and only at selected locations over the
model domain. They cannot provide the regular hour-average type data
37
-------
needed to conduct a complete evaluation. Moreover, the National Ambient
Air Quality Standards (NAAOS) are written, for ambient surface level concen-
trations. The ROM evaluation will emphasize the model's ability to predict
these concentrations.
There are many diagnostic tests that could be applied to elucidate the
phenomena inherent in the ambient data. At a basic level we are interested
in observing whether the air quality data, particularly 03, define a plume
downwind of major source areas. The ROM evaluation in an episodic mode is
predicated upon the impacts of such plumes on downwind locations. A careful
spatial analysis of the available data should reveal the extent of these
plumes and whether they are broad enough to be resolved on the 18.5 km grid
scales of the ROM. For receptors outside the influence of strong hydrocar-
bon sources a simple chemical check of the photostationary equilibrium
between NO, N02» and 03 can be made under daylight conditions. This test is
particularly useful for flagging monitors (either NOX or 03) that may be
malfunctioning for some reason. The expected interrelationship among these
species outside of hydrocarbon influence provides an independent gauge on
the adequacy of the monitored values. This test is of limited use however
for low NO concentrations when instrument sensitivity becomes a factor.
Another diagnostic procedure is a check on the NMHC/NOX ratio in the air
quality data. For a given time of day, this ratio can be studied over all
monitoring stations. Values in the vicinity of urban source areas can be
indicators of the potential smog reactivity of the pollutant mixture.
A seemingly recent data technique of spatial (and temporal) analysis,
called Kriging (after the mining engineer D. H. Krige), should prove to be
especially useful in the analysis of the network data from fixed monitors
and in transforming that data into a spatially smoothed form most relevant
for comparison with ROM predictions. Kriging is based on an application of
the statistical properties of random functions. It has come from geostatis-
tics (Delhomme, 1978; David, 1977), but for spatial analysis it is the same
as structure function analysis (Gandin, 1970) and also has application to
time-series analysis.
38
-------
Kriging was developed as a method for optimal interpolation, accounting
not only for the geometry of data values -(distance), but also for distance-
dependent structure in the data related to spatial covariability (spatial
correlation). It is a method for interpolating fields of data that mini-
mizes the error variance of the interpolated point. Thus Kriging provides
a means to estimate the interpolation error of points and/or regions in a
data field, under certain assumptions of random behavior of the data. In
developing such estimates, Kriging provides a means to estimate inherent
sampling error together with variability due to subregional effects that are
evident from the data. Thus, Kriging provides a means to estimate a more
regionally representative pollutant field for comparison with grid-averaged
pollutant predictions. It also provides a means of estimating the extent
of influence of spatial covariation and the degree to which such a range of
influence is anisotropic. Hence, Kriging provides an estimate of the accu-
racy or uncertainty of both the monitoring network values and the interpo-
lated values.
Kriging should be a useful tool in the comparison of model output data
sets with a validation data set that contains inherent sampling (and measure-
ment) error. As an interpolation method, it can be used to develop point
or spatial estimates from network values, together with an estimate of the
associated uncertainty, corresponding to ROM grids. Using these uncertainty
estimates and the theory of standard normal distributions, the statistical
significance of differences between individual ROM realizations and interpo-
lated validation data can be assessed. Areas of significant model error
can be identified and contour maps of regions of significant error can be
produced to investigate the spatial characteristics of model errors. The
uncertainty estimates can also be used in more complex characterizations of
spatial and temporal model errors.
Through the use of Kriging, together with co-located monitoring data,
it may also be possible to establish the expected variance in the air
quality measurements at monitoring stations due to measurement error. The
variance estimated for a particular monitoring site for a given time period,
39
-------
such as 1 hour, is a combination of instrument variability and small scale
chemical effects plus other subregional effects. ROM layer 0 63 predic-
tions include an estimate of these subgrid scale chemical effects. Thus
some comparisons should be possible between the expected variance as de-
•
termined from selective Kriging of the network data and that from the ROM
predictions.
In addition to analyzing the air quality data, the emissions and
meteorological data bases can be used in diagnostic testing. The source
emissions distribution across the ROM domain in the Northeast U.S. applica-
tion is not at all homogeneous. Major urban areas emit large amounts of
nitrogen oxides and hydrocarbons from a distribution of sources. Large
single point sources, such as fossil-fuel electric power plants, are also
scattered throughout the ROM domain in urban and nonurban areas. Emissions
from these point and area sources are transported downwind according to the
existing flow regime. Determining the trajectories along which these
emissions travel can be an interesting and useful exercise. These traject-
ories should provide guidance on the location of actual and simulated
plumes from these sources. A subset of major point and urban area sources
can be chosen from all sources within the ROM domain. For each gridded
wind flow realization during an episode a forward trajectory is tracked for
the duration of the episode. This can be done for each realization in the
family of possible realizations. Comparisons can be made among the computed
trajectories for a given source area. If the trajectories show good agree-
ment in location among all realizations the model should provide a clearly
defined predicted plume center! ine in the air quality simulation with which
to compare against the observed plume data. Trajectories showing spatial
disparities among the realizations may indicate a less precise predicted
plume location. In this case some trajectories may intercept other major
downwind source areas and other trajectories may not, leading to possible
large differences in the computed air quality values in the predicted
plume. In any case, the envelope of trajectories defined by the family of
realizations should provide the boundaries of the territory where the
actual source plume is located. When these boundaries are very
40
-------
far apart, the utility of the air quality predictions for short temporal
periods may not be great, although the result shows the spatial area that
is "at risk" over a longer period of time for the same set of synoptic
conditions. This type of information is important for the model evaluation
because it quantifies some of the limits of predictability for the ROM as
used on a specific data base.
EVALUATION BY INDIVIDUAL REALIZATION
The simulation of any given episode with the ROM will involve a finite
number (~ 10) of realizations based on possible flow fields. We can regard
any one of these realizations as determining the "real" outcome, while ex-
cluding all the other realizations. This is a deterministic perspective
where we search for the best predictive capability of the model by eval-
uating each realization independently. In this case ROM predictions can be
directly compared with monitoring station observations. Because of the
strong influence of initial conditions on the modeled results, the first 6
to 12 hours of simulation should probably be excluded from the evaluation.
This will enable the ROM to begin producing concentration fields based on
the source emissions input at the point the evaluation begins. Since model
initialization occurs at 1200h (noon) the first results subject to evalua-
tion should be those of OOOOh the next day, with the evaluation proceeding
in increments of 24 single hour periods.
The focus of this evaluation will be on the accuracy of the ROM predic-
tions of hourly 03 concentrations, including the daily spatial and temporal
patterns of the predictions, and the accuracy of the magnitude of the peak
prediction for each day. The first point is important in interpreting the
results, establishing confidence intervals, and diagnosing errors, while the
analysis of 03 peaks is at the heart of the regulatory use of the ROM.
41
-------
Dennis et al . (1983) give a thorough discussion of various measures of
model performance. The choice of particular measures depends upon the
attributes of performance deemed important for a particular model and data
base. For the ROM evaluation by individual realization, the performance
measures appropriate for determining the accuracy of the hourly predictions
•
are the bias, gross error, noise, and variance, complemented by graphical
analysis of temporal and spatial behavior. For evaluating the peak predic-
tions, all of these measures as well as a linear regression of predicted and
observed values are appropriate. Estimated model bias is the mean of the
residual or difference (d) between observed and predicted values, or,
d - (C0 - Cp) = C0 - Cp.
If the bias is significantly different from zero, it indicates a systematic
tendency in the model's predictions with respect to timing, spatial location,
or magnitude of the peaks. Gross error, a measure of model inaccuracy, can
be estimated by the average absolute deviation in the residual concentrations,
W- (ICo - Cp|).
The estimated noise is the variance in the residuals, or,
u n - 1
Both the noise and bias are components of the gross error in the model. The
noise level is generally high for model results that are strictly paired with
observations by time and location because a prediction that is off by a few
42
-------
hours or grid cells can greatly affect the noise level. The noise may be a
more meaningful performance measure when-it is based on less strict pairings,
such as the observed and predicted daily maxima.
Graphical analysis is also a key element in model evaluation. Informative
plots may be constructed from the following elements:
(1) Frequency histograms of C0, Cp, and d,
(2) Time series plots of C0 and Cp,
(3) Scatterplots of d vs. time or C0,
(4) Boxplots of d within natural categories, such as monitoring site,
hour or day of observation,
(5) Spatial contours of C0, Cp, and d fields at time of maximum C0, and
(6) Plots of wind flow trajectories.
For each ROM realization we can apply the relevant performance measures
to hourly values of 03 for all receptor locations in the domain, strictly
pairing the observed value with the predicted. This should be done both for
predictions in layer 0 (shallow surface layer) and layer 1 (lower portion of
mixed layer) in order to determine the effect of simulating the near-surface
physical and chemical processes on the model's predictive performance. Model
predictions of CO might also be evaluated in this manner to test the perfor-
mance on a nearly inert pollutant species, although regionally representative
CO monitoring data are not commonly available. Also, for all receptor loca-
tions the estimated variance in the hourly 03 observations from the Kriging
data analysis can be compared with the layer 0 predicted variance in the ROM
63 values for the same time and location. Receptor sites for the comparison
of variances might be stratified according to urban and non-urban locations
to separate the microscale chemical effects expected only in the urban areas
from the total variance. Thus comparisons can be made for the hourly average
values of 03 and its estimated variance between the predictions and observations,
The step outlined above in the ROM evaluation should provide estimates
of the model bias for each realization as well as possible signs of any
43
-------
systematic errors in the model results. The-analysis of these results can be
done for each individual day of the simulated episodes (up to 24 strictly
matched pairs of C0 and Cp at each receptor location) as well as cumulatively
for all days in an episode, and for all episodes. Furthermore it is sometimes
useful to form subgroups of the hourly values on which statistical evaluation
can be performed. For example, model performance can be judged for 03 above
a certain threshold level, such as 0.12 ppm which is the current NAAQS value.
Another subgrouping might stratify the hourly values as to time of day such
as 06 - lOh, 10 - 14h, and 14 - 18h. There may be other natural dividing
lines for subgroups as well. The evaluation by subgroup can often detect
performance characteristics unique to that subgroup which might be masked or
averaged out in the full set.
The next step in the evaluation procedure is the analysis of the peak or
maximum values; a step of great interest to the regulatory community since
the peak values are those that dictate emissions control steps. We concentrate
on peak hourly 03 values in this discussion. There are 2 sublevels of analysis
for peak values: local and global. At the local perspective we choose the
daily maximum observed hourly value at each receptor site. This can be paired
with the prediction at the same site and time as well as the daily maximum
hourly prediction at the site for any time. Both pairings are useful
comparisons. If there is a systematic bias of the predicted peak value in
time of day it should appear in the comparison of results from these local
maxima analysis methods.
From a global perspective the observed daily maximum hourly concentration
over all sites within a specified area of the ROM domain is determined and is
paired with the corresponding predicted value in one of three ways: (1) with
the predicted peak at the same site as the observed, (2) with the predicted
peak among all sites within the specified area, and (3) with the predicted
peak from any grid cell within the specified area of the domain. As these
pairings become less restrictive (from 1 to 3) we should expect a tendency
44
-------
away from underprediction in the model performance. With all other factors
being equal the third pairing should lead to overprediction because the
observations are constrained to the monitoring sites only, a smaller sample
of points than all the grid cells in the area. However, the simulation of
high 03 episodes sometimes leads to a tendency toward underprediction of 63 in
all 3 pairings because studies of model performance are often done for days
where high 63 concentrations have been observed and days with lower observed
concentrations have been excluded (Dennis et al., 1983).
The specified areas of the ROM domain to be used in the global analysis
of 03 maxima should be chosen based on some common characteristics that tie
the area together. Examples might include political jurisdictions such as
individual states, airsheds for air quality regulation based on population,
source emissions, terrain distribution, or large downwind sectors from
major source areas.
Another aspect of the ROM evaluation by individual realization that
might be more difficult to quantify is the comparison of the spatial patterns
in the predicted and observed concentration fields. Graphical contouring of
these fields as well as the residual concentration field can subjectively
provide an estimate of the model performance in this area. An objective
graphical technique described by Siege! (1982) compares the shapes of two
data sets and demonstrates the motions necessary to overlay one on the
other with the best possible fit without altering the intrinsic shape of
either field. There is a potential utility for this type of analysis in the
ROM evaluation. Each individual simulated realization is a possible analog
of the actual concentration field. If coherent plumes of higher concentration
are seen emanating from large source areas in both the simulated and predicted
fields, we can apply the objective graphical technique to compare these
plume shapes. The predicted plume can then be translated and/or rotated as
necessary to provide the best fit to the observed plume. This will, in
effect, normalize the predicted field for errors caused solely by the specified
wind field. After this plume rotation has been performed on all possible
45
-------
areas within the model domain the statistical analysis of predicted and
observed 03 concentrations at receptor srtes is again made to see how the
plume reon'entations improved the effective model performance. This step
will be most useful in comparing the fields surrounding the maximum hourly
observed 63 concentration to that of the predicted 03 within the same plume
area. This comparison can be made for each realization for each day of simu-
lation.
Figure 12 shows a structural diagram of the possible steps in the ROM
evaluation by individual realization. The structure diagram is a hierarchy
of operations including sequences (shown by horizontally colinear boxes),
iterations (denoted by an asterisk, *), and selections (denoted by an open
circle, 0). The exact path we choose to take during the evaluation will be
guided by the results obtained at intermediate steps along the way (feedback
mechanism) as well as the available resources at hand. The diagram encompasses
the complete set of evaluation exercises of which all or a subset of those
shown will actually be performed.
EVALUATION OVER ALL REALIZATIONS
•
The distinguishing aspect of the ROM setting it apart from most existing
air quality models is its ability to determine an expected range within
which the actual measured concentration should fall. This ability is achieved
through the assimilation of prediction data from all ROM realizations for a
given episode. From this perspective all the realizations must be considered
concurrently. At any given receptor location the ROM results generate a
frequency, or probability distribution of expected concentrations.
This section explores potential avenues of evaluating this probabilistic
type of model prediction. It is clear that in this vein there is no longer a
one to one correspondence between prediction and observation. At any receptor
location and time a single observation value (and possible variance, based on
46
-------
EVALUATION BY INDIVIDUAL REALIZATION. PAGE 1
SHORT-TERM
ANALYSIS
EPISODE
ANALYSS
PROCESS
INDIVIDUAL
DATA VALUES
STATISrCAL
ANALYSJS
Of DATA
1
RECEPTOR
ANALYSIS
-OUR
ANALYSIS
.,,
MAXIMUM,
VALUE
ANALYSIS
SPATIAL
PATTERN
ANAL"SS
'
TOTAL
DAY
ANALYSIS
Figure 12. Structure diagram of the ROM evaluation by individual
realization.
47
-------
EVALUATION BY INDIVIDUAL REALIZATION. PAGE 2
PROCESS
CONC
VARIANCE
SAVE 1
MAXIMUM 1
V*LUE j
c0U,t)
Cp(x.t)
vs
SKIP STEP0
'f NOT
LAYER 0
PERFORM °
SfEP IF
LAYER 0
ajtx.t)
=i|(x.t)
Figure 12. (continued)
48
-------
EVALUATION BY INDIVIDUAL REALIZATION. PAGE 3
AM»LVZE
CCNC -
ALL HOURS
PROCESS
VARIANCE
SKIP SfEP°
IF NOr
(JVYEF) 0
SO 5
IF L»YE
£D 0
fl 0
Figure 12. (continued)
49
-------
EVALUATION BY INDIVIDUAL REALIZATION. PAGE 4
NO °
ANAl^SlS
FOR SPECIES
00 c
ANAir&S
FOR SPECIES
AMM.YZE
CONC - ALL
RECEPTORS
P90CESS
VARIANCE
Figure 12. (continued)
50
-------
EVALUATION BY INDIVIDUAL REALIZATION. PAGE 5
NO 01
ANA.TSIS 1
FOR SPECIES |
DO °
ANALYSIS
fCR SPtCES
LOCAL !
PERSPECTIVE |
i
RECEPTOR
1
DO WIND
TRAJECTORY
DEFINE
GLOBAL
AREAS
TERATW
COMPARE
MAX
VALUES
GLOBAL
A9EA
COMPARE
MAX
VALUES
Figure 12. (continued)
51
-------
EVALUATION BY INDIVIDUAL REALIZATION. PAGE 6
«(XA) ,s
Figure 12. (continued)
52
-------
EVALUATION BY INDIVIDUAL REALIZATION, PAGE 7
1
CHOOSE HO'.f
OF CBS
CONG kWX
r
SPATIAL
PATTERN
ANALrSlS
I
NO °
ANALYSIS
FO* SPECIES j
~\
00 °
ANALYSIS
FOR SPECIES
i
ill!
CEFINE
DOMAIN
SECMENrs
SEGMENf
STEP
!
DOMAIN '
SEGMENTS
CONSTRUCT j RE-DO
REORIES'EO EVALUA'lOS
BEST flELO [ STEPS |
1 1
9£CE°TCR 1 MAXIMUM I
ANALYSIS VA^UE i
ANALYSIS '
Figure 12. (continued)
53
-------
EVALUATION BY INDIVIDUAL REALIZATION, PAGE 8
TOTAL DAT
•40 °
ANALYSIS
FOR SPECiS
1
KXJRIY
VALUES
00 3
ANALYSIS
COR SPE&ES
1
[ MAXIMUM
I VALUES
ANALYZE TOR
ALL HQUBS,
^CEPTCRS
PROCESS
VARIANCE
ANALYZE MAX
CONC - ALL
RECE°:ORS
1 ANALYZE MAX
t ALL GLOBAL
AREAS
SKIP STEP0
\r NOT
LAYER 0
PERFORM °
srtP if
WE' Q
Figure 12. (continued)
54
-------
EVALUATION BY INDIVIDUAL REALIZATION, PAGE 9
I FIND SINGLE II NO °
I SESr REAU- I 4NairS:S
I ZA'lON '09 FOR SPECIES
1 . .
LAV£R ! 1XIF.R 1
UNALTZE s**» NAIY^E w*x
-ALL RtCEPr cose OVCR
I AN? DATS I ICLOBAI. AREA)
Figure 12. (concluded)
55
-------
a Kriging analysis) is available. At the coarsest level of evaluation we
can examine the resultant concentration distribution for a particular receptor
and time, and judge its utility by its shape. For example, Figure 13 shows two
hypothetical predicted concentration distributions at a receptor for a 1-hour
period. The predicted concentration values for this receptor from the 10
realizations, along with their assigned probabilities of occurrence based on
the wind field realization probabilities, define a concentration distribution.
The distribution shown on the top is rather broad and flat, indicating that
concentration values from C^ to C2 are nearly equally likely to occur. If,
for example, the species is 03 and the range of C^ to €3 is from 0.05 to 0.20
ppm, the model result is not very useful in an absolute sense. However, if
the distribution changes uniformly with changes in emissions, then the
prediction may prove useful to the regulator in a relative sense. One
contributing factor to the possible prediction of a broad distribution such
as this one is the location of the surface and upper air wind monitoring
stations in relation to the receptor location. For example, consider the
case of a receptor located in the same grid cell as a wind monitoring site.
The method of wind field interpolation for the ROM forces the calculated
fields to conform to the measured values at the monitoring site locations.
Therefore the closer a receptor is to a wind monitoring site, the smaller the
differences will be among the various wind estimates for the grid cell
containing the receptor in the multiple wind realizations, and hence the
smaller the differences will be among the predicted concentrations from the
realizations. The prediction of a broad distribution, such as the exaggerated
one at the top of Figure 13, is possible for receptor sites that are well
removed from wind monitoring stations where the constraints of conformity to
the observations are not so powerful.
The distribution shown at the bottom of Figure 13 shows a narrower
peaked distribution. The expected range here should be more useful in an
absolute sense to the regulator and to the evaluation procedure. At a minimum
we can determine whether the observed concentration falls within the predicted
distribution and whether it is within 1 standard deviation (a) of the mean.
56
-------
o
z
o
LU
DC
C2
CONCENTRATION
CJ
Z
O
LU
ac
CONCENTRATION
Figure 13. Hypothetical predicted concentration distributions
from the ROM.
57
-------
If the observation falls outside of or in the tail of the distribution, the
model prediction is clearly in error. The concentration variance calculated
from a Kriging analysis of the observed concentrations is not exactly comparable
to the variance of the predicted distribution because the former is based on
instrument variability and subgrid scale variability while the latter is based
on expected variability from uncertainty in the flow field. New techniques,
such as combined temporal and spatial Kriging, might produce more comparable
estimates of the distribution.
From an evaluation standpoint we need a distribution of observed concen-
trations in order to perform objective statistical comparisons. There must
be an orderly grouping of receptor stations to fulfill this requirement. It
is not immediately obvious how such a grouping is achieved. One criterion
for the grouping is that there should be no significant difference in the
predicted concentration distribution among the receptor members of the group.
We need a single predicted distribution with which to compare the observed
distribution. Therefore the receptors in the group must have essentially the
same prediction. This criterion can be used to decide if we have chosen a
proper grouping of receptors. There is also no assurance that a proper
grouping can be found from the receptor network set we will be working with.
Figure 14 shows an example from an early run of the 1st generation ROM
of a box plot of the predicted 03 concentration distribution from 69 layer 1
grid cells containing mostly urban land-use types. This plot was made from a
single realization. If the shape and magnitude of this distribution remained
constant over all realizations the grouping would be valid. This plot is
shown by way of example only, for it is now thought that a grouping based on
land-use type alone will not highlight the physical phenomena occurring in
the regional scale model.
Intuitively one can imagine plumes of 03 streaming downwind from major
source areas in the model domain. The paths of these plumes will intersect
receptor locations. One can hypothesize a grouping of receptors based on
58
-------
-
-
-
—
-
—
-
_
-
1 * i •
TRIBUTION IN LAYER 1 OVER 69 URBAN CELLS MAXIMUM VALUE
3rd QUARTILE
MEDIAN
1st QUARTILE
MINIMUM VALUE
CONCENTRATION DIS
1 ^ 1
'"*• '"i" ' ' 'i
i 1 . , i 1
i ' ' ' '
' l_.,._l_33 «•
i L. 1. .„,.,! '
i — 4 1 1 •
nJ n-r-n
,' =
, L 1 1
1 * l
' , i 1 j t
1 ( , . +
r i ' i i
I i i • I
1 U_ 1 1 1
1 1 J — J '
I 1 i < (
1 ^ t — ._.';„. ..T..J ^ t
i 1 n 1
i ( n — i
• ' ' T
e
o
a
o
o
o
S
s
o
o
«T
O
0
o
CM
a
o
o
CM
0
o
CM
CM
0
s
CM
0
o
oo
a
o
CO
o
0
o
a
CM
CM
cn
cc
C/3
O)
O
s-
cn
c
•i— CO
-o rs
a> cr,
s- rs
<4- S-
o o
M-
a. s-
i cu
x >,
O (O
co •—
O)
S-
cn
LD o in
T-' ,-' d
tudd'(0tx) (\|OllVail\l3DI\l03
59
-------
considerations of plume transport such that the members of the group lie
within the downwind extent of such plumes as close as possible to the center-
line and at comparable distances downwind of the respective source areas from
which the plumes originate. The receptors should be chosen within plumes
•
that have maintained their integrity and have not yet coming!ed with other
major plumes. Figure 15 shows a conceptualized idea of how these receptors
might lie in relation to several major urban source areas and their plumes.
Any given plume should meander about a downwind receptor over the set of
realizations. However the predicted concentration distribution formed from
all realizations will have this meandering effect built in and should be
common to all receptors in the grouping. This grouping can then be tested
against the criterion described above.
Other considerations are necessary for this grouping also. For example,
this discussion assumes that the major source emission areas have similar
characteristics. In reality this is not true, so the predicted 03 concentrations
must be normalized in some manner by the source strengths of NOX and hydrocarbons
and the ratio between them (THC/NOX). Also, the time period of interest must
be specified for a receptor. This might correspond to the hour during which
the maximum 03 concentration was observed at a receptor during the day, or
else an hour by hour 63 comparison could be attempted between predicted and
observed distributions.
If it is then determined that the predicted concentration distribution
formed from all realizations is invariant over the set of receptors in the
grouping, this distribution can then be compared to the distribution formed
from the observed concentrations at the receptors. Should the predicted and
observed concentration distributions show significant differences it would
imply that some aspect of the chemistry or physics in the model is incomplete
or incorrect. If, however, the distributions are similar the model results
would be verified for this particular set of source-receptor relationships
under the given set of meteorological conditions, but the general validity of
the physics would not necessarily be verified.
60
-------
c/1
c
o
re
N
(C
ta
O
O
S-
o
4->
Q.
O)
O
O)
I
a;
o
s_
3
O
LO
OJ
s-
C7I
61
-------
The test outlined above for the evaluatipn of the ROM over all realizations
involves searching for a set of receptors that meet the criterion previously
described. Dimensionally we are holding time fixed (within a single day
period) and choosing a set of spatially varied receptors. There is an
alternative path for evaluating the ROM over all realizations that involves
focusing on only a single receptor at a time and allowing the time dimension
to vary. This procedure would involve choosing a receptor within the downwind
extent of an urban plume and examining the model results on a set of days
that have essentially the same meteorological conditions and emissions
functions. This last point implies a criterion that would assure invariance
in the predicted concentration distribution from one selected day to the next
at the chosen receptor. The observed distribution is then formed from the
measured concentrations at the receptor on the selected days. A favorable
comparison between the predicted and observed concentration distributions
would imply that the ROM is mimicking the stochastic variability seen in the
atmosphere. The problem with this evaluation method is that a very large
data base spanning many weeks is required in order to select days that have
virtually identical meteorology. In theory the approach is sound, but
practically our application data base for the Northeast U.S. has only a
limited temporal time span that is not sufficient to carry out this method of
evaluation. For now this path of evaluation will be abandoned, although in
the future it may be pursued given sufficient resources and a more tractable
method of approach.
Thus the recommended method for the ROM evaluation over all realizations
is the original one described here. The set of receptors used from day to
day would vary depending on the wind direction. Preliminary trajectory
analysis is necessary to determine whether there are potential source-receptor
combinations available to carry out this analysis. Should it prove to be
feasible the analysis should be made for ROM results from layers 0 and 1 for
03 for the hour of the observed maximum and also possibly for all hours where
the average concentration within the class of receptors is above a certain
threshold value. Figure 16 is a structure diagram summarizing the steps
62
-------
EVAL OVER ALL REALIZATIONS
SRCE-RCPT0
COMBINATIONS
NOT FOUND
SRCE-RCPT0
COMBNAT10NS
FOUND
SKIP
REST
OF STEP
DO REST
OF STEP
SPECIES
*
ANALYSIS °
NOT
REQUIRED
PERFORM °
ANALYSIS
A
Figure 16. Structure diagram of the ROM evaluation over all
realizations.
63
-------
EVALUATION OVER ALL REALIZATIONS, PAGE 2
A
T
PERFORM
ANALYSIS
HOUR OF
MAX OBS
CONC
B
HOUR
ITERATION
1
*
HOUR
0
SKIP
SVtP
0
DO
STEP
B
"cm.
Figure 16. (continued)
64
-------
EVALUATION OVER ALL REALIZATIONS, PAGE 3
COMPARE FRQ
DIST AMONG
REAUZATONS
REST
OF
STEP
SKIP STEP °
IF PRO DIST
NOT ALIKE
DO STEP °
IF PRO DIST
ARE ALIKE
FORM DBS C
DIST AMONG
RECEPTORS
COMPARE
DBS, PRED
DISTRIBUTNS
Figure 16. (concluded)
65
-------
that are possible in the ROM evaluation overall realizations.
Various objective statistical tests may be used to compare the charac-
teristics of the observed and predicted distributions. The standard t-test
for comparison of means and the F-test for comparison of variances can be
made. In addition, the Kolmogorov-Smirnov (K-S) test, a powerful "goodness-
of-fit" test between two cumulative distributions can be made. The K-S test
is valid for small sample size and it is conservative in that it minimizes
the chances of making a Type II error (i.e., accepting the hypothesis that
two samples have been drawn from the same population when, in fact, they are
not).
FURTHER TESTING
An ideal experiment for evaluating the transport aspects of the ROM
including the predictive uncertainty quantified in the multiple realization
concept would involve the release of a controlled amount of passive tracer
from a single source that is subsequently sampled in a dense receptor network
on regional scales. A recent field study, the Cross-Appalachian Tracer
Experiment (CAPTEX) conducted in 1983 was such a study (Clark et al., 1984).
An inert trace gas was released on 5 separate occasions from a point source
in Dayton, OH and on 2 other occasions from a Sudbury, Ontario point source.
A dense sampling network was established in arcs at downwind distances of 300
to 1100 km from the Ohio release point in approximately 100 km intervals with
additional sites to supplement the sampling from the Canadian release. In
addition to the surface measurements, upper air wind data and aircraft sampling
data are available as part of the CAPTEX data base.
An experiment could be designed using the ROM to simulate one or more of
the tracer releases and subsequent plume evolution over the eastern U.S. and
Canada. Some of the questions that might be investigated in this type of
experiment include:
66
-------
(1) Do the model's various realizations define the envelope in which the
real plume is observed?
(2) What is the temporal and spatial variability in the real plume?
These questions, and others, might be answered more definitively in a controlled
emission experiment such as this one. The current configuration of the ROM
domain for the Northeast U.S. application would have to be enlarged on the
west and north sides for this experiment to bring in the source locations.
Alternatively, the sources could be specified as flux boundary conditions on
the current domain, but this would involve making assumptions on the behavior
of the gas from the source point to the model boundary, another potential
source or error. Current resource levels preclude the undertaking of this
modeling experiment with a new data base, although future ROM applications,
when feasible, should include this one.
A model as complex as the ROM contains dozens of input parameters for
which the model might be sensitive to slight errors in their specified values.
A thorough sensitivity study of the ROM is beyond the scope of the project
outlined here. However there are certain aspects that are relevant to the
evaluation study and the present use of the model.
The choice of the horizontal grid cell size for the model domain is a
significant decision for it determines the resolution on which regulators
must base decisions. It also determines the resource level needed to operate
the ROM in terms of computational time and memory requirements and the
resolution and quantity of input data. For the ROM the grid-cell size will
also affect the degree of subgrid scale chemistry that is accounted for in
layer 0. One interesting sensitivity test that is recommended is to test the
effect of horizontal grid size on model results. Since the current resolution
of approximately 18.5 km was the resolution used to construct the source
emissions inventory a smaller grid size would be difficult to use. However
larger cells of double size (~37 km) or triple size (~55.5 km) might
67
-------
be used in this test. For instance, would the predicted concentration in one •
large 37 km cell be equal to the average.of the four 18.5 km cells contained
within it? Discrepancies between these results might be attributed to subgrid
scale processes.
Another test might be a ROM run to determine the contribution of individual
large source areas on downwind receptors. This would involve an advection/
dispersion run with the chemical kinetics turned off. Hypothetical tracers
corresponding to the N species in the chemistry algorithm would be released
from the N largest emission source areas, one species to an area. Then the
concentration at downwind receptor sites could be analyzed for the component
contributions from each source area. Results would be valid only for inert
species, since the effects from chemical reactions could change the relative
contributions based on advection/dispersion alone. Other tests might also be
possible, depending upon the available resources.
68
-------
SECTION 5 .
SUMMARY AND CONCLUSIONS
The U.S. EPA Regional Oxidant Model and NECRMP data base have been
described here. The model incorporates a comprehensive description of the
physical and chemical processes thought to be important to tropospheric 03
production on 1000 km scales. The data base employed for the first application
of the ROM was collected during the summers of 1979 and 1980 in the Northeast
U.S. It contains meteorological and air quality data from regular monitoring
networks as well as enhanced networks or special field project measurements
made during that period.
The evaluation procedure that will be used to determine the ROM performance
on this data base has been outlined. A number of episodes will be simulated
from the period July 23 through August 16, 1980 for which performance statistics
will be developed. The evaluation of any given day within an episode will
proceed in 2 distinct stages. The first stage will focus on model performance
for an individual model realization, irrespective of all other realizations.
Model realizations for a given day are functions of the possible flow fields
that existed for the day. The second stage will attempt to evaluate model
performance using the full probabilistic abilities of the ROM that consider
all realizations concurrently. The focus of the evaluation will be on 03.
Whether other species shall be considered and the exact pathway through the
evaluation study will be determined by the resources available at the time.
In conclusion it should be worthwhile to review the intended uses of the
ROM. The model was originally conceived as a way to provide the initial and
boundary conditions required by urban scale models. The modeling system
would be structured as shown in Figure 17. The ROM would first be run using
projected future year emissions to establish initial and boundary conditions
for the urban model. The ROM inputs would also include worst-case meteorology.
69
-------
METEOROLOGY
REGIONAL OXIDANT MODEL
URBAN OXIOANT MODEL
MAXIMUM OZONE CONCENTRATION
MAXIMUM OZONE
EQUAL TO
ESIRED VALU
CREATE NEW
REGIONAL
EMISSIONS
IS THE
NEW REGIONAL
EMISSION INVENTORY
THE SAME AS THE
PREVIOUS?
ANALYSIS
COMPLETE
Figure 17. Emission reduction analysis using the ROM and urban
scale models.
70
-------
Model results would be used to specify the initial concentrations and boundary
concentrations for each urban area. The urban worst-case meteorology would
be a subset of the regional worst-case scenario. The emissions input into
the urban model would be adjusted until the model predictions showed, that the
reduction in peak ozone concentration would be sufficient to insure that the
oxidant standards are not violated.
After the emission reductions are determined for all urban areas a new
regional emissions data base, describing the future year emissions assuming
the implementation of control strategies, can be compared with the original
regional emissions. If the emissions differ greatly, in either time or space,
then the ROM would be run again to determine new urban initial and boundary
concentrations. If the new initial and boundary conditions are substantially
different, then the urban models would be run again to refine the emission
reduction requirements. The procedure would be iterative.
Another use of the ROM has also been proposed. It would be used to
predict ozone concentrations in urban and rural areas directly. The direct
use of the ROM is illustrated in Figure 18. Emissions are adjusted and
model predictions are made until the desired ozone reductions are achieved in
all areas in the region. Also along these lines there is an interest in
using the model to predict longer-term regional average concentrations.
For example, a growing-season 63 average is appropriate for determining
potential crop losses in the agricultural belts of the midwest and central
plains areas of the U.S. Logistical problems exist in a long-term ROM
application, such as assembling the data base needed to make such a run or
choosing a subset of days to represent an entire season.
Knowledge of how the ROM will be used is important in determining the
objectives of the model evaluation. All of the potential uses of the model
described here will be considered when the evaluation is performed and the
results are analyzed. Explicit emissions scenario testing, however, is not
a part of this study, but may be conducted in a future study.
71
-------
REGIONAL
EMISSIONS
INITIAL
CONDITIONS
BOUNDARY
CONDITIONS
REGIONAL OXIDANT MODEL
MAXIMUM OZONE CONCENTRATIONS
IS MAXIMUM
OZONE EQUAL TO
DESIRED VALUE?
ANALYSIS
COMPLETE
Figure 18. Emission reduction analysis using the ROM alone.
72
-------
REFERENCES
Clark, T. L., G. J. Ferber, J. L. Heffter, and R. R. Draxler, 1984: Cross-
Appalachian tracer experiment (CAPTEX '83). In Proceedings of the
4th Joint Conference on Applications of Air Pollution Meteorology,
16-19 October 1984, Portland, OR, American Meteorological Society,
Boston, MA, 158-161.
David, M., 1977: Geostatistical Ore Reserve Estimation, Elsevier Scientific
Publishing Co., Amsterdam, 364 pages.'~
Delhomme, J. P., 1978: Kriging in the hydrosciences. Advances in Water
Resources, 1, 251-266.
Dennis, R. L., M. W. Downton, and R. S. Keil, 1983: Evaluation of performance
measures for an urban photochemical model. EPA-450/4-83-021, U.S.
Environmental Protection Agency, Research Triangle Park, NC.
Dennis, R. L. and J. S. Irwin, 1985: Current views of model performance
evaluation. In Proceedings of the DOE/AMS Model Evaluation Workshop
(23-26 October 1984, Kiawah,SC), Vol. I: Participants and Invited Speakers
Papers, A.M. Weber and A.J. Garret, eds, E.I. DuPont de Nemours and Co.,
Savannah River Laboratory, Aiken, SC 29808.
EPA, 1982: Northeast corridor regional modeling project - annual emission
inventory compilation and formatting, Volume 1: Project approach.
EPA-450/4-82-013a, U.S. Environmental Protection Agency, Research Triangle
Park, NC.
Freas, W. P., 1983: Northeast corridor regional modeling project-data base
description. EPA-450/4-83-015a, U.S. Environmental Protection Agency,
Research Triangle Park, NC.
Gandin, L. S., 1970: The Planning of Meteorological Station Networks, WMO,
No. 265. TP. 149, Geneva, Switzerland, 35 pp.
Lamb, R. G., 1983: A regional scale (1000 km) model of photochemical air
pollution. Part 1 - Theoretical formulation. EPA-600/3-83-035, U.S.
Environmental Protection Agency, Research Triangle Park, NC.
Lamb, R. G., 1984: A regional scale (1000 km) model of photochemical air
pollution. Part 2 - Input processor network design. EPA-600/3-84-085,
U.S. Environmental Protection Agency, Research Triangle Park, NC.
Possiel, N. C. and W. P. Freas, 1982: Northeast corridor regional modeling
project - description of the 1980 urban field studies. EPA-450/4-82-018,
U.S. Environmental Protection Agency, Research Triangle Park, NC.
Sellars, F. M., T. E. Fitzgerald, J. M. Lennon, L. J. Maiocco, N. M. Monzione,
and D. R. Neal, 1985: National Acid Precipitation Assessment Program
emission inventory allocation factors. EPA Contract Report 68-02-3698,
U. S. Environmental Protection Agency, Research Triangle Park, NC
(March 1985).
73
-------
Siege!, A. F., 1982: Geometric data analysis: an interactive graphics program
for shape comparison. In Modern Data Analysis, R. L. Launer and
A. F. Siege!, eds., Academic Press, New-York, 103-122.
74
-------
APPENDI-X A
SURFACE MONITORING SITES
This appendix contains a listing of all surface monitoring sites in the
special NECRMP field studies and the SAROAD networks in the Northeast U.S.
taking 1-h air quality measurements of 03 during 1980.
75
-------
>-
C3
f-lss^O*O^*DKl^h*O'a'>O
CMrHi-IWKli-lrHCJrHrHOJKlCJ
CJCJrHrHrHKIKI CJrHKIrH
Kl Kl CJ
ooLnoLftooLftoo-3'uiuioocjo.*ocjuioooir)o.o •*• irt o co o
Aiaj.j'inajr-.a^cjr^.rHKio^rHrHrHcjf'-.ocjooiriuioajLft.yknrH cj -*• o u"i KI
oKior in
ooooooooooooooooooooooooooooo ooooo
UJ I
ID —1 •
O O
i-lrHrHOrHOr^rHrHKIi-lr-irHOOCJ
•* U1 .*• *• *«« »** fcrf X.* Ml %>• -XI ^ Ul XJ %XJ «•*
rHrHrHOrHOr~rHrHKIi-lrirHOOCJrHrHi-lrHrHrHrHi-li-lrHrHrHCJ CJrHrHOKI
OOOOOOOOOOOOOOOOOOOOOOOOOOOOO OOOOO
oooooooooooooooo
ZZ
ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
3 3 3 3 3 3 O
U CJ U U U U U
WHIrHrHHirHrHUJ
H-l-H-Hr-t-t-Q:
CJOUCJCJ(JU<
UJLUUJLUUJUJUJ2
zzzzzzzz
< <
3 3
< <
&. on
< <
CK *:
< CJ
ii X
U U
CJ U
ii ii X
U U CJ
* ^
u u
^ ^ ^
CJ U U
^ ^
CJ CJ
^^
OOOOOOOLLJLUUJUJ
UOCJUUUOQOQQ
UJ HI UJ UJ UJ UJ UJ UJ UJ UJ UJ LU til UJ
O O O
U
M HI <
U. U- I
o: tt
M M 2
< < UJ
U. U. Z
So
U
O
-I O
UJ CX
HI O
U. U.
a H-
HI o:
O
u o o
CJ U
a
-i x z
LU UJ UJ
M V) >
LU LU <
X -J X
CJ O
: HI M uj
- z
o o
u u
I O (O (O
I CJ < <
, CJ U
i Z 3 3
: uj uj uj
o o
CJ CJ
UJ Ul
_l -I
S!S
CJ CJ
3 3
UJ UJ
O
x8
UJ
in a
tn >-
o
U
O >-
U <
3
UJ O
o
U
O -I
U -I
UJ
UJ CO
o
u o
u
I ^
:yS
I CJ O
o < < <
o o o o o
U U U CJ CJ
o z 8 z z z z
CJ O O O O O
co ui co co in if>
oz a Z a a: CK a:
LU UJ HI UJ UJ IU UJ
cj a ^ u. u. u. u.
a: z u. u. u. LU u.
II I II I Q It I it I It I l! I
uj in
I— a:
I— UJ
UJ U.
>- u.
< UJ
u. -)
o
u
o o o o
U CJ U O
u
o
QL
Sv
UJ
111 It I III Mj
-1 -J -J -I O =
KH-Kr-O O>-OUJ
tot^yiio QCJ^H-—!
x 3 5 z a
c a uj o uj a a:
n a HI H- > ce O
: rtOu.uj>-Su.x-ixu,(-Q:cjcjucjuj<
O CO 00 "' "^ ' * ^* " ^ '" 'rt '
HI Z
a: <
03
o
CJ
Ul
UJ
o o o o <
C9 CJ O O CJ U.
OOQ^O I UJ UJ
CJCJ05ZCJ=ZZZZ-J-J
CJLJO OOOOO-J-J
uj •*. ujtowoiyicnuiH-HiHi
H-cjZQD:ZD:a:Q:D;C5>>
r-OMOUJHIUJUJUJUJLItnift
>3>-Zu.X-JXU,(— CKCJCJUCJUJr~UJCJOC£;QSiiUU
iZQ^ujQfhHO3i^K-'>2223v^xO*-J^Zx— >*ZO^D^ZL^Ui>
: rH ^x
CJ CJ CJ O CM O O
rH rH rH O rH O O
O Ifl O O O CO O
xo fx» o Ki cj r^» r^*
O rH rH Kl xT xT Lfl
O O O O O O O
o o o o o o o
rHrHrHrHrHrHinininrHKIinrHLTlrHrHrHLflinrHinKlrHCxIinrHrHrH
oooooooooooooooooooooooooooo
LULuU.LUU.LU«tLU±COCOOOO*OxDCOOCOxfxrcJCJCJCJOCOCO
f**OiHOrHi-liHrHCJOC\JincO^OiHinxQxQr^COxr>0sOCT'K!KIKt
OrHrHOOOOOOOOOOOOrHi-lrHrHi-lrHrHr-lrHrHCJCJCJ
r^^%rx,cOcocO
-------
VI
Q
**
CM iH rH rH CM. CM, rO r-l rH K» rH rH rH CM
M U"> Kl'3'^'Cf'NyrOOOOO rHrHrHrHCMOOOi-HrHrHOrHOl/lKirOOO'd'roroOO
UI S OOOOOOOOOO OOOOOOOOOOrHOOOOOOOOOOOOO
CO -J —
O O
I-l rHrHCMrHrHrHrH OrHrHrHrHOrHOr-lrHrHrHrHr-lrHrHrHCMrHi-lrHrHrHi-H
O OOOOOOO OOOOOOOOOOOOOOOOOOOOOOOO
ooooooooooooooooooooooooooooooooooo
ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
ro f> ro
>->•>•>->->->->->-
(JUULJUUUUU
oaaaaaaaaaaaaaaa
ZZZZZZZZZZZZZZZZ
<«:<<<<<<<<<<<<<<
ZZZZZZzzZrHMrHwMMrHMttattttttttaaciittttaDrttttttCKCi:
UIUJUJUJUJUJUJUJUJ<<-
UI UJ rv
a uj o u> CK >•• uj
:g§uu,i3§
I rH 10 -J O M U 10
08
u
•f
uj ct:
r- r- r- H :
co CQ :
o
o
o
u uj
C£
a o
cr E
OM
o
O O i
U U I
•z. to i
uj cr i
2*£ UI I
U CO I
< z :
a: uj i
u —i •
O o
u cj
i in
! o z :
i x:
: o i
3
o o
< < :
a a. i
co
re <
UJ M
i— M
in _i :
UJ UJ
u ui
z a.
rH <
3 U
o
u
o o o o
u u u u
OOO-I-J-J-I OO
UUUUIU1UJUJ O U
Q a o a
r->-l— ZZ^ZUIUJUJUJ
._OOO3333tECKQ:n:
3_iuj-
u -i
a:
UJ
—I
-J Q
tH 'Z.
> < tfl Q
in -J 3 o
>- a: u o
o
a:
in
z
CJ CJ
o o
N in
CJ O
o o
o o
•o -o
^ cr
oj CM
tri <-i
o a
u. u.
i-l UI
10 o
00
o o
o o
u. u.
PJ ^p
o o
o o
o o
•a 1-4
j- f^
Kl 10
CO CO
l/> tfl
o o
u. u.
CM i-H
O O
o o
o o
o m
rH CM i
1/1 m i/i
o o o o o
u. u. u. u. u.
rH CM rH *3" m
O O O O O
o o o o o
o Is* o r*- to
I/I i-H
o o
U. O
•O i-H
o o
o o
O CO
o o
O O
O O
Kl VJ-
O
-------
in
o
iH rH Kl rH CJ rH rH
K> IO IO K> K» CJ rH
M CO
UJ E
X
UJ I
to -j <
O CD
oouitfoinoocjoinoooooLnin o"ff> vfiminoopnioooinoinooM
O-OfOIOOtfintfrHrOCJCJtOCJinOOjr~incJi-HfOfOfOin-oa»OOCjrOajtftfO
inrH-OOtfOrHOrHrHOOOiHiHrHrHCJOIOOOO.-lrHOtfa-O^vOinr-tO'O
OOOOOOtfOOOOOrHOOOGOOOrHOGGOOOGOOOOOOO
i-lrHrHCJiHrHfOrHiHrHrHrHIOtfrHrHrHr^rHrHrHrHiHrHrHrHrHrHrHrHrHrHCJi-HrH
OOOOOOrHOOOOOOOOOOOOOOOOOOOOOOOOOOOO
32232333323303202333323233:302333:332
I O O rH f11* O
i o in tf cj tf
ooooooooooooooooooooooooooooooooooo
LU
§
ZZZZZZZZZZZZOZZOZZZZZZZZZZZOZZZZZZZ
r->ejtfr>-oina)cvj>o^ff>rHoioo-3>oc>-o^
•friHtfrOrHtOtfrHOtfOtfOIOOOintfrHOOtfr-IOCJCJKIOtfKI :> .XI CJ O CJ
OTWuiOTinininininin
r-r-r-l-r-(-r-r-l-
t- r- I- t-
VI
§i
< <
_j _j
>• >-
a. a.
<. <
2: i:
11.
-J -J .
ct re i
999
"": «e <
—i -j
•>•>-:
or o: i
UJ iij
tr> tf)
3 13
X X
U (J
< <
tn in
co cn
XX
(J U
in in
in in
LU LU
-J M
o o o o o o
u u u u u u
uxxxxcrn: o
LULULULULULUOOO
^COlOCOinh-r^CJCJO in
-JUIUJUJUJCOCO OUJ
O—I-J—1-JLULULULULUUCO
n Q n Q O U CJ Z Z Z UJ
O
u
UJ O
LU U
in
LU t—
Z Z
LU UJ
ID ^
O
U
o
UI
M
in
o
u
U —I O
rH UU H-
cc co co
ui z o:
O UI UJ
in til rr>
o: ce <
u. ID X
o
U O
5
•<
a
n:Q
OO:
U-
M^
>U
O
O
CE
o
i:a
I- Z o; a ce -i
—I C£ 2 LU UJ < -J
M>-3OI-r-a.r-r-rH
>o:cOr-(nco rn>-H>
n:3coc:LJLuZOOX
UJEl^LUUUUJCECiLU
nQ_»r-Ct:C;-Jr-r-CO
O3LU IT)
CJ Kt >O
O O r-«
O O O
O O O
ro oj j-
O i-t CvJ
O O O
o o o
u. u. u.
p-t CJ rH
O O O
o o o
cj o en
r» co to
ro O O3
I** CO CO
i-H *H rH
C-.I N C-J
OJ to ro
CJ CJ CJ CJ
rH CJ K» tf in
r» r- (^ i^ r^
78
-------
fH iH K) rO IO CJ IO I
z uiooojoounoooor-OO~>~>~'
OUUUOUUUUCJOU32222222222322232222233
MMMIHMMMIHMMIHrHLUUJUJLULUUJLULULUUJUJUJUJUJLUUJUJLUUJUJUJLULU
mECErEErEiisEzzzzzzzzzzzzzzzzzzzzzzz
o o
u u
Sy
ill
o
u
O 111
O U I-
U I-
O O
U U
zz
OO
o
3 U
<
Z D
OO
oo:
O
O
L)
o
<
x(j
OM
OO
uu
o
O
u>-
<
Z
oo
O
U
O
o>-
a:
XUJ
O
uo
O
ZMOMX
o 8
u o o
u u o o
x o z o o
UJ |- (- o < O
to uj uj —i o zz
ujcO(no:Q:L>-JLu
-J o: o: uj LU o z
QLUUJCJCQUJLUO
OHZTO;2Z'HU_O
wooiusotZJo:
£(Oinx:uuj5O<
^^UJ^U^LUUJ ^i »-tujjOtHa:cQZQ:f——ID:-JX
OH--UJ
•^o:-i>-r:i:Q-E:inuj^: zo:33 r- i
to to
to to i
oooooooooo
U.U-U.OU.U.U.U-U.U.
OO
OO
OO
i-HOO
OOO
OOO
OO
OO
OO
oo
U.U.
tO^-l
OO
OO
OO
oooooooo
O
O
O
o o o o o o
n u u n u f i u n ii
OOOJfHfHOJOJOJOJf^
ooooooooo
ooooooooo
ooooooooo
COOO-OO-OIOCOW
Nj-iAroo
-------
IOIOIOK1 (OK1IOCJCJCM rH IO IO Kl IO Kl CM IO lOIOKlrHCJ rH CJ CM rH Kl
M )
UJ Z
O-rOOOOOOOOOOOOOOOOOOOOOOOOOr-HrHOOOrHOC)
S_J «,
O
lHlflrHr-IOrHrH.OincJ.OCMlOrHrHrHr.(r-lrHrHOOrHi-HrHi-HCMrHrHCMr-ICMCJr-lrH
OOOOGOOOOrHOOOOOOOOGOOOOOOOOOOOOOOOO
Ul
g
333333333:33332333333333:2333:33:333323
ooooooooooooooooooooooooooooooooooo
ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
ooo
oooooooooooooooooooooooo
*" *">°'>"oooooooo
333333333333333333333333333MrHMMMrHi-irH
UIUJU1U1U1UJUJUJUJUJUJUJU1UJUJUJUJUJLJLUU1UJU1UJUJU1UJXXXXXXXX
ZZZZZZZZZZZZZZZZZZZZZZZZZZZOOOOOOOO
o
u
§>->->•>->->- tfl
r-r-t-r-r-r- -J Ul
rH W MWrHrHMM-JrH >• i£
r- -J :<: uuuuuuo<( in ox ao
^ —J O O CJ O O Uu O CL UJ Q£ ^ O
o O^^t^incj^isiiiiiiii^^i U UJ 3 3 *C *t u.j 3 1Z CJ 3* O Q£ Of D£
(O ^ UJ CD 1C 31 3Z Z ^1 2Z SI Z Z! 21 !Z 21 O CU Q^ Q£ Ul (rt tn (O ^
to
M
O -J
u a
: uj ui
oo t<
UZ rli
: z ui < o M M
^1.ujiu^.^-^l.UJ«^^Jf-ir-i
(JZKOOUJajUJr-UO
w>-rHt!:Q:-'cj:azzz
r-•
Kl Kl Kl Kl 1
Kl Kl IO Kl I
^ Kl
o o
o o
o o
-O CM
o so
1/1 Ul
Kl Kl
Kl Kl
O O O
U. U- U-
r-l VT Kl
O O O
O O O
o o o
CO vO C\l
-o r^ o
in in -o
Kl Kl Kl
Kl K) Kl
O O
-> u.
rH rH
O rH
O O
O O
CM CJ
in *o
Kl Kl
Kl IO
O O O
U. U. U.
Kl Kl Kl
O O
X U.
O rH
CM O
O O
O O
•£ CO
o o
o o
rH rH CM rH rH
O O O O O
X X X X X
CO CM rH O O
O O O rH CM
O O O O O
O O O O O
-O CO O CM CJ
Kl in O CJ CM
O O rH rH rH
Kl Kl Kl Kl K> Kl Kl
80
-------
a
Kll-lrHrHCJCjrJrOrOfl CM rH KI rH
rOKIKIrHKICaCUKIKI
U>
in
OOOOrHOO OOOOOOOOOOrHOOOOrHrHrHOOOOOrHOOO
1U
g_J .
19
-
< a
o o
E X
o
zz
o o
>- oo
o; o o
uj uj o O Z Z
—' ^ ^" O O
-J-i-JcQOOZZr-t-
o
O O
o
CO
io
oo
OO
zi:i-H -I -J O H < <
oo
oo
o to
UJ UJ
o
8
UJ
I-
M
CO
o
o
o t-
Si
cc a:
< uj
o
o o
Ul r-
^* Z
O O
O O
o o o o o
CO W
pi
= 2
—i —i
o o
o o
10 I-
o o <
ooo
Z :
o :t i
o o i
UJ
UJ
Z <
O M
i- a:
>->-
< -j
a uj
< r- r-
FH -J _J
o: M M
5§§
UJ X X
Sz
< <
CQ CD
UJ O
1-4 O f
a a .
O -J
OC —I O O M
uj i-l O O >
i: > uj o z
O tO CD UJ UJ O O
CDLU^-Ji£CQQD
I—Zr-D3c:3ujuj
ZWQ;UJ<(UJ—J—i
o-
O O
O O
CO CD
UJ
O O
U O
>- >-
UJ UJ
CD 3
Ul UJ
a
o
a:
to
z
o o
X CD
I/I iH
Kl O
O O
O O
CJ ^O
C\J CJ
rH i-H
rH iH K> •-<
O O O O
X X U. u.
cj ^r i-t O
Csj O O O
o o o o
o o o o
»0 CO CO CJ
-o o a* r*-
»H r-t iH CJ
ro
I KI KI I
CJ rH
o o
X X
O OJ
iH O
o o
o o
CJ O
r» ^
CJ K»
-o -o
i-4 KI
oo
U. »-4
CJ »H
oo
00
oo
O CJ
in co
to n
rH r-4 rH rH CJ
O O O O O
U. X CD tD CD
rH o O
ro ro KI
o o
M X
CJ *O
rH O
o o
o o
CJ O
•3- -O
-JD sO
•O -£>
ro ro
o o
X X
rH ro
CO O
o o
o o
O CO
»o cj
r-4 rH in 1/1 m
O O O O O
M CD ~> I ->
r^ r^. f^s o «*H
o xo NO co oo
o o o o o
o o o o o
,00000
r^ r-t iH (-* r-1
r-- o o o o
i fn ro fo m ro
rHCJCJCJCJCJCJCJCJCJCJCJ
81
-------
i-i
* CM CM i-t iH CM
ooooooooooooooooooooooooooooooooooo
ZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ
IU
§
OOOOOOOOrHrHOO-OOOCVIOOOOOOOOOOOOO
I -f IO •» I
UJ ''' iu iu "' "' UJ iti 14J ii I 114 111 jjj ii i i£j n i ii i nj tij 111 111 uj j^j ti i
0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.
O O I-
O U U Q.
u z:
o: o; <
V
•z. o
UJ O
Si
«<
o
r- UJ i
to D: :
O.
i:O
.
u o
LJ
u
D:2:
w
O ui
< o r-
M a
': uj <
i aj —i
O O O O O O O
o uuuuuuu
o u
o <<<<<<<
Z rH M M rH M W rH
>-o xxxxxxx
on- a. o. a. a a. o. a
uj a. ~~
§§
(J) x
z o;
o o
z: z
UUIUJUJUJUJUJUJ
a o a a a o o
>-<<<<<<<
D:_I-J_J—i_i_j—j
OirHrHrHMrHrHt-l
Hj""""^"*""""""
a.
O O
u u
M M
X X
a. a.
a o
23
i I—* i—i i
: x x x x x
. o. a. a. a. a.
o
u
o
UJ
OJ
uj x
—I M U
-------
a
CJ M IO
rO IO
rOiHiHiHiHCJiHr-ICJiHrOrOiHiH
.
CJ Kl rO
CJ CJ
M in
u i:
UJ
o u
OOOOOOOOOOOO OOOOOOOOOOOCJO i-HOOOCJOOO
ocjcjcjcjinint^oocjcjin-ororomcjcjoroinocj^to \nr-cjo M
o a.
< uj
x x
o
°o
X O
<
< z:'
u. in :
85
M
O J O
I— IH O
O. 3
- ^ a.
O
U
o
UJ UJ
o o:
= S
z <
UJ X
uj uj
a -J
U) <
o
CQ
tn
a. uj
«£ K
01 in
UJ UJ
X X
< <
u. U-
a: o:
D: o
UJ I—
a D.
O O
U U
a: o
UJ U
> M
o o;
o
u
O E
u a: <
Uo
to u
a: <
UJ UJ
z to z
K O
O < h-
u IH tn
z uj
Z M -I
M ID
o: Z
a w
Z Z
O O
Z c, 13
w
D; X
13 3
X X
in in
< <
3 3
§
o
a:
in
z
o o
U) u.
co o>
o o
o o
o o
•O in
O iH OJ
ooo
ooo
CJ Cvl CJ
iH O O
OJ O O
oo
o o
CJCJCJCJCJOJCJCJCJCUCJCJCJCJCJCJOJCJCJCJCJOJCJOJCJCJOJCJOJCJCJCJCJCJCJ
83
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
. REPORT NO.
2.
3. RECIPIENT'S ACCESSION NO.
4. TITLE AND SUBTITLE
EPA REGIONAL OXIDANT MODEL: DESCRIPTION AND
EVALUATION PLAN
6. REPORT DATE
6. PERFORMING ORGANIZATION CODE
. AUTHOR(S)
Kenneth L. Schere and Allan J. Fabrick
8. PERFORMING ORGANIZATION REPORT NO.
L PERFORMING ORGANIZATION NAME ANO AOORESS
10. PROGRAM ELEMENT NO.
CDWA1A/02-4021 (FY-85)
(same as 12.)
11. CONTRACT/GRANT NO.
12. SPONSORING AGENCY NAME ANO AOORESS
ATMOSPHERIC SCIENCES RESEARCH LABORATORY - RTP, NC
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
RESEARCH TRIANGLE PARK, NC 27711
13. TYPE OF REPORT ANO PERIOD COVERED
In-house
14. SPONSORING AGENCY CODE
EPA/600/09
15. SUPPLEMENTARY NOTES
6. ABSTRACT
The U.S. EPA Regional Oxidant Model (ROM) and NEROS data base are described. The
model incorporates a comprehensive description of the physical and chemical processes
thought to be important to tropospheric 03 production on 1000 km scales. The data base
employed for the first application of the ROM was collected during the summers of 1979
and 1980 in the Northeast U.S. It contains meteorological and air quality data from
regular monitoring networks and from enhanced networks or special field project
measurements made during that period. The evaluation procedure that will be used to
determine the ROM performance on this data base is outlined. A number of episodes will
be simulated from the period July 23 through August 16, 1980, for which performance
statistics will be developed. The evaluation of any given day within an episode will
proceed in two distinct stages. The first stage will focus on model performance for
an individual model realization, irrespective of all other realizations. Model real-
izations for a given day are functions of the possible flow fields that existed for the
day. The second stage will attempt to evaluate model performance using the full
probabilistic abilities of the ROM that consider all realizations concurrently. The
focus of the evaluation will be on Og. The exact pathway through the evaluation study
will be determined by the resources available at the time.
7.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
b.lDENTIFIERS/OPEN ENDED TERMS
COSATi Field/Group
18. DISTRIBUTION STATEMENT
RELEASE TO PUBLIC
19. SECURITY CLASS (This Report)
UNCLASSIFIED
21. NO. OF PAGES
20. SECURITY CLASS (This page I
UNCLASSIFIED
22. PRICE
EP* Form 2220-l(R»v. 4-77) PREVIOUS EDITION is OBSOLETE
-------