United States
Environmental Protection
Agency
Environmental Sciences
Research Laboratory
Research Triangle Park NC 27711
Research and Development
EPA-600/S3-84-074 Aug. 1984
&EPA Project Summary
Air Quality Models
Pertaining to Paniculate Matter
S. A. Batterman, J. A. Fay, D. Golomb, and J. Gruhl
This report describes an evaluation of
the Pollution Episodic Model, an urban-
scale dispersion model that incorporates
deposition, gravitational settling, and
linear transformation processes into the
predecessor model, the Texas Episodic
Model. A sensitivity analysis of the
model was performed, which included
the effects of deposition, gravitational
settling, and receptor grid size.
Recommendations are made to
improve the performance and flexibility
of the model.
Pollution Episodic Model was applied
to a source inventory of the Philadelphia
area to provide a preliminary estimate
of source apportionment. Pollution
Episodic Model modeling employed
both hypothetical and actual
meteorology. Results indicate that area
source emissions dominate total
suspended paniculate, sulfur dioxide,
and sulfate concentrations at urban
receptors. A large fraction of the
inhalable particles may arrive from
distant sources.
This report also contains an overview
of receptor models (RMs) used for the
source apportionment of aerosols.
Some diagnostic procedures for RMs
are evaluated with a synthetic data set.
Described are receptor model tradeoffs
and protocols and possible hybrid
dispersion/receptor models. Issues re-
garding the intercomparison of source
apportionments from receptor and dis-
persion models are highlighted with
reference to the 1982 Philadelphia
study.
This Project Summary was developed
by EPA's Environmental Sciences
Research Laboratory, Research Tri-
angle Park, NC, to announce key
findings of the research project that is
fully documented in a separate report of
the same title (see Project Report order-
ing information at back).
Introduction
The principal objective of this work was
to assess the two fundamental
approaches to source apportionment of
aerosol concentrations: source-oriented
dispersion modeling (DM) and receptor
modeling (RM). We aimed to define the
strengths, limitations, areas of
applicability, and possible protocols for
the use of these methods in the
regulatory context. The specific
objectives included the following:
(1) Assessment of the new Pollution
Episodic Model (PEM), a dispersion
model that incorporates deposition,
settling and transformation
processes into the standard
Gaussian plume dispersion
algorithm. Pollution Episodic Model
is an extension of the Texas
Episodic Model (TEM-8).This report
contains a discussion of the verifi-
cation, applicability, and limitations
of PEM. Suggestions are given to
improve the versatility of the model.
(2) Review and general comments
about receptor models. Part of this
project involved a review of the
models and literature that make up
the current understanding of recep-
tor techniques, as well as our
experience with receptor models
and similar statistical problems.
The current status and capabilities
of RM are discussed, with attention
to the potential uses of RM in the
regulatory process and the
-------
application of diagnostic tools to
RM that might be part of the
modeling protocol.
(3) Discussion of potential hybrid dis-
persion-receptor models. There are
a number of ways to combine
source and rejeptor models.
Currently, it is not clear that hybrid
models are feasible and desirable.
However, it is clear that
contributions from each approach
can be made to improve the other
technique in certain circumstances.
Recommendations are made
regarding the development and
application of hybrid models in
intermodel comparisons.
(4) Comparison of DM and RM in the
Philadelphia Study. In July and
August of 1982, data were
collected in the Philadelphia area,
in part to compare source
apportionments from DM and RM
on the same, real data. Ground
rules and certain specif ic tasks were
required to make that comparison
as meaningful as possible.
Guidelines and a discussion of
issues for this comparison are
presented. Preliminary modeling
results that used PEM and a source
inventory compiled for the
Philadelphia area are given.
Evaluation of the Pollution
Episodic Model
The PEM model successfully incorpo-
rates simplified deposition, gravitational
settling, and linear transformation
processes. Pollution Episodic Model
simultaneously calculates concentration
and deposition over an urban-scale area
in a rectangularly gridded receptor
network.
With respect to deposition, PEM
provides greater flexibility and some
increased realism over comparable
models. For example, the Industrial
Source Code (ISC) model simulates
gravitational settling by tilting the
(normally horizontal) plume centerline;
deposition is modeled by the partial
reflection of the source contribution.
Settling and deposition processes alter
the distribution of mass within the plume.
The ISC model's treatment is not
altogether satisfactory because it does
not model the modified distribution.
Consequently, the ISC model tends to
overpredict concentrations near the
source and underpredict concentrations
at large distances.
The PEM source code was checked for
obvious errors or omissions. No mistakes
were found. The source code was then
compiled on the IBM FORTRAN V compil-
er. Only one "block data" per program is
permissible with this compiler. The
source code was modified accordingly.
The PEM was used with simple source
and meteorological conditions in each
stability category. Results were
compared to the TEM-8 model. This
showed good agreement, provided that
the TEM time step was specified to be 10
minutes. Both models use the same P-G-
T dispersion coefficients, but TEM in-
creases these coefficients for calculating
one hour averages. (The basic averaging
period in PEM is one hour; in TEM, it's 10
minutes.) Maximum differences between
the two models were under a few per-
centage points. However, under the most
stable categories at relatively long
downwind distances, maximum
differences could range up to about 30
percent. These differences occurred be-
cause TEM uses "look-up" tables for dis-
persion coefficients at discrete distances,
while PEM uses piece-wise approxima-
tions. Although both methods should be
sufficiently accurate for model applica-
tions, the piece-wise approximations in
PEM may provide more consistent
results. Finally, analyses were performed
to verify the model's response an
sensitivity to changes in settling and
deposition velocities, transformation rates,
and area source treatment.
The PEM model was found to be easy to
use (even without a user's guide).
However, the versatility and convenience
of the PEM program could be improved.
Some of the recommendations in this
section result from our experience with
the model; others are proposed in view of
expected applications. Implementation of
the minor changes to the model (e.g.,
bookkeeping operations, additional
outputs, and new user options) would
facilitate sensitivity analysis and source
apportionments. Other changes might
increase the realism of the model. The
somewhat increased complexity,
memory storage, and computer time
required by the model would not be
important considerations to most model
users.
Of the recommended changes, the
most important improvements that
increase model credibility are those
concerning area source modeling. For
sulfate modeling, hourly variation of
transformation rates and elimination of
cutoff distances for point and area source
calculations are the most relevant. For
TSP modeling, source-specific
specification of particle size or deposition
and settling velocity are important. The
development of a long-term version of
PEM for seasonal or annual particulate
concentrations would complement
existing air quality models. Most of the
remaining recommendations simply
provide greater convenience for model
users.
Receptor Models
Receptor models (RMs) are procedures
that use observed aerosol characteristics
to identify and quantify the sources of
ambient air pollutants. The aerosol char-
acteristics most frequently measured
include chemical and elemental mass,
optical properties, and particle size
distribution. Less frequently measured
characteristics include isotope ratios,
organic and inorganic compounds, and
crystalline structure. Unlike dispersion
models, RMs do not require meteorologi-
cal or source data. Source profiles may be
used, but are not necessary.
In general, RMs use the following
assumptions:
(1) The total aerosol mass at a receptor
is a linear sum of the aerosol
contributions from individual
sources. In general, there cannot be
any transformation of the aerosol
from time of emissions at the
source through atmospheric trans-
port to collection and ultimate
analysis.
(2) Characteristics of the aerosols are a
linear sum of the aerosol
characteristics from individual
sources (e.g., elemental ratios are
assumed to be constant between
source and receptor).
(3) Source apportionment is possible
only for those source classes that
have identifying characteristics. No
unique characteristic is needed,
just a unique combination of
characteristics.
(4) Chemical mass balance models
require quantification of source
characteristics (e.g., elemental
ratios). Multivariate models require
less precise descriptions, but
patterns of characteristics must be
recognizable. Sources with un-
known or highly variable character-
istics may not be distinguishable.
-------
(5) For regression models, the number
of characteristics must exceed the
number of source classes, and
errors (residuals) should be
normally distributed and
uncorrelated. For multivariate
models, the number of filters must
be sufficient for the degrees of
freedom required for the number of
filter characteristics and sources
used
Deviations from these assumptions will
degrade the validity of the receptor model.
The magnitude of the deviations that typi-
ally occur in RM applications is not known
at the present time. Also unknown is the
susceptibility of RMs to such deviations.
There are obvious situations where
assumptions are violated and RM
applications may not be useful.
Hybrid Approaches
There are several possibilities for
hybrid dispersion/receptor models.
Hybrid models use a receptor model
approach in conjunction with source
emission rates, source locations, and/or
dispersion information. Such models
conceivably might provide a more
accurate and flexible means of source
apportionment. They may also help
reconcile differences between source
and receptor models. All data, including
meteorology, source emissions and
profiles, and filter characteristics are
used in coupled models. Coupled models
intimately combine receptor and
dispersion approaches, and thus may be
the most complicated of the hybrid
approaches. Coupled models might be
classified by their primary orientation,
around either dispersion or receptor
approaches. For receptor-oriented hybrid
models, outputs of DMs might be treated
stochastically. Chemical Mass Balance
(CMB) RMs might then be extended by
either including DM predictions as priors
in Bayesian optimizations, or by using DM
predictions of source contributions as
coefficient weights in weighted
optimizations. Analogous methods are
possible for multivariate RMs Here,
preprocessed meteorological data might
become new variables in a factor analysis
or multiple regression model. Association
of particular meteorological patterns with
source class signatures may identify the
likely direction and distance of contribut-
ing sources. Alternately, source
contributions predicted by DMs may
become variables in factor analyses,
along with the usual filter characteristics.
Associations of predicted source
contributions with the respective source
profiles may indicate areas of agreement
between the models.
The second general type of coupled
model is based around dispersion
modeling. The (deterministic) inputs or
parameters of DMs are considered as
stochastic variables (with estimated
distributions). Optimization is used to
match DM predictions to observed
concentrations or receptor-determined
apportionments, at the same time
maximizing the probability of the
stochastic variables in the DM.
Some major issues can be identified
relative to the use of receptor models.
Currently, there is a substantial amount
of imposed "intelligence" that is
introduced into RM procedures by
experienced practitioners. Present RM
studies are generally custom-designed,
including selection of source signatures,
filters, and analysis. In the regulatory
context, some of the discretionary
elements of RM might have to be
removed to reduce the chances for
misuses of the models.
The variation in source profiles (over
time and space) and the deviation from
the RM assumptions can be determined
from source or near-field measurements,
or perhaps derived by using wind
trajectory analysis. Synthetic data sets
with realistic error structures and
compositions for various aerosol regimes
(e.g., rural western vs. urban eastern)
may be used to construct RM protocols.
Hybrid models require more data than
either approach alone Such models may
be cost-effective if source inventories and
meteorological data are available. If
hybrid models permit significantly im-
proved performance in terms of accuracy
and flexibility of source apportionment,
their expense and complexity may be
justified in other circumstances. At
present, hybrid models have not been
demonstrated and critically evaluated.
Conclusions and
Recommendations
A new dispersion model, the Pollution
Episodic Model (PEM) was found to
successfully incorporate deposition,
gravitational settling, and linear
transformation processes to the
predecessor model, the Texas Episodic
Model. Thus, PEM should permit greater
realism in urban scale modeling.
However, some improvements to area
source calculations seem warranted.
Application of PEM to a source
inventory for the Philadelphia area and a
sensitivity analysis indicated the
following:
(1) Area source emissions may
dominate TSP, sulfur dioxide and
sulfate concentrations at receptors
located within an urban scale.
Therefore, source inventory data
for area sources, including source
strengths, operating schedules,
microinventories (around receptor
sites), and the degree and method
of aggregation of small sources
may be critical to accurate
dispersion modeling. Distant
sources warrant less attention.
(2) A large fraction of the observed
inhalable particle mass, particu-
larly sulfate, probably comes from
medium- and long-range transport
and not from local sources.
Preliminary modeling has shown
that sulfate contributions from local
sources are generally only a few
micrograms per cubic meter,
compared with measured values
which average about 24 jjg/m3.
However, the model examined a
relatively short period, and the
source inventory is known to be
very approximate.
(3) Particle and gaseous concentra-
tions from middle to far field sources
(greater than 10 km) may be sensi-
tive to deposition velocities, with
greater effects at larger distances.
For these sources, gravitational
settling is relatively unimportant.
Gravitational settling may be im-
portant only for sources located in
the vicinity of monitors that emit
large particles with high settling
velocities.
(4) Sulfate concentrations from distant
sources are sensitive to
transformation rate, especially at
low wind speeds.
Our review of the literature and
experience with receptor models
indicated the following:
(1) The temporal and spatial variation
of source profiles, and the sensitiv-
ity of estimated apportionment to
such variation is largely unknown.
The amount of data needed for
representative and useful results is
also poorly defined.
(2) Generally, RM studies have been
custom-designed, including
selection of sources signatures,
filters and analysis. A high degree
-------
of subjectivity may be involved in
the interpretation and use of data
and models.
(3) Each type of receptor model is
prone to certain failures and limita-
tions. Chemical mass balance RMs
are subject to problems of collinear-
ity and influential points.
Diagnostic procedures can be used
to determine whether these
problems exist; remedial
procedures may be able to minimize
their effects. Standardized
protocols may be required to ensure
meaningful results and promote
appropriate uses of RMs.
Both dispersion and receptor models
have useful attributes for the source
apportionment of aerosols. Dispersion
models are predictive and diagnostic.
Receptor models are primarily
interpretive. Standardized uses of recep-
tor approaches are possible; however,
their applicability and limitations need to
be defined.
Hybrid models, which combine aspects
from both dispersion and receptor
approaches, may have applications to
many air pollution problems, including
apportionment of ambient concentra-
tions of criteria and hazardous pollutants,
visibility impairment, and acid deposition.
However, at present, hybrid models are
poorly developed and defined. The de-
velopment and validation of hybrid models
will require an extensive data set for
various conditions and localities.
S. A. Batter man, J. A. Fay, D. Golomb, and J. Gruhl are with the Massachusetts
Institute of Technology, Cambridge, MA 02139.
Jack Shreffler is the EPA Project Officer (see below).
The complete report, entitled "Air Quality Models Pertaining to Paniculate
Matter," (Order No. PB 84-210 939; Cost: $11.50, subject to change) will be
available only from:
National Technical Information Service
5285 Port Royal Road
Springfield, VA22161
Telephone: 703-487-4650
The EPA Project Officer can be contacted at:
Environmental Sciences Research Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, NC 27711
it U S GOVERNMENT PRINTING OFFICE, 1984 — 759-015/7770
United States
Environmental Protection
Agency
Center for Environmental Research
Information
Cincinnati OH 45268
Official Business
Penalty for Private Use $300
<• ' ' - ' I- , 1 L C ') O i) C4
------- |