FEBRUARY 1Q86
EVALUATION OF THE PEM-2 USING THE
1982 PHILADELPHIA AEROSOL FIELD STUDY DATA BASE
ATMOSPHERIC SCIENCES RESEARCH LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
RESEARCH TRIANGLE PARK, NORTH CAROLINA 27711
-------
EVALUATION OF THE PEM-2 USING THE
1982 PHILADELPHIA AEROSOL FIELD STUDY DATA BASE
by
Jia-Yeong Ku and K. Shankar Rao
Atmospheric Turbulence and Diffusion Division
National Oceanic and Atmospheric Administration
Oak Ridge, Tennessee 37830
IAG-DW13930021-01-2
Project Officer
James M. Godowitch
Meteorology and Assessment Division
Atmospheric Sciences Research Laboratory
Research Triangle Park, North Carolina 27711
ATMOSPHERIC SCIENCES RESEARCH LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
RESEARCH TRIANGLE PARK, NORTH CAROLINA 27711
-------
NOTICE
The information in this document has been funded in part by
the United States Environmental Protection Agency under Interagency
Agreement IAG-DW13930021 to the Atmospheric Turbulence and Diffusion
Division of the National Oceanic and Atmospheric Administration.
It has been subject to the Agency's peer and adminstrative review,
and it has been approved for publication as an EPA document.
Mention of trade names or commercial products does not constitute
endorsement or recommendation for use.
ii
-------
ABSTRACT
The Pollution Episodic Model Version 2 (PEM-2) is an urban-scale
model capable of predicting short term ground-level concentrations and
deposition fluxes of one or two gaseous or particulate pollutants at
multiple receptors. The two pollutants may be chemically coupled
through a first-order chemical transformation. PEM-2 is intended for
urban particulate modeling, and for studies of the atmospheric trans-
port, transformation, and deposition of pollutants to assess the impact
of existing or new sources or source modifications on air quality.
This report describes an evaluation of the PEM-2 using Phila-
delphia Aerosol Field Study (PAFS) data for 29 days in the summer of
1982. The model's performance is judged by comparing the calculated
12-hour and 24-hour average concentrations with the corresponding
observed values for four pollutant species, namely, S02, sulfate, fine
and coarse total mass. The calculated and observed diurnal variations
of hourly SC>2 concentrations at each of the six PAFS stations are
also compared. A first-order chemical transformation of S02 to sulfate
is considered in the calculations in addition to the direct emission
and dry deposition of all four pollutant species. The model%domain,
covering 80 km x 80 km with 32 x 32 grid cells, includes 300'point
sources and 289 area sources in the Philadelphia urban area. Hourly
meteorological and emission data are used as inputs to the model.
Statistical tests for evaluation of model performance include
standard measures of differences and correlation between observations
and calculations paired in space and time. For each pollutant, scat-
terplots of calculated 12-hour average concentrations and differences
versus observed concentrations are presented; a linear regression line
is determined and evaluation statistics are tabulated. Additional
plots and tables, examining the model performance for daytime,
nighttime, and daily mean concentrations (averaged over all PAFS
stations) are given. The diurnal variations of S02 concentrations
(averaged over the evaluation days) are also compared at each station.
The emphasis in this evaluation is on particulate species for
which the model performs well; the variations of 12 and 24-hour mean
particulate concentrations over the evaluation period are simulated
closely. These results, however, should be interpreted with caution
since the background concentrations far exceed the urban source
contributions to the particulate concentrations in Philadelphia. The
model performance for SC>2 is better during daytime than at nighttime,
and generally better at the suburban stations than at downtown
stations which are impacted heavily by the urban area sources.
The work described in this report was performed by NOAA's Atmos-
pheric Turbulence and Diffusion Division in partial fulfillment of an
Interagency Agreement with the U.S. Environmental Protection Agency.
This work, covering the period January 1984 to June 1985, was completed
as of June 30, 1985.
iii
-------
CONTENTS
Abstract ................ . ...................................
Figures [[[ vi
Tables [[[ ix
Acknowledgements ............................................. x
1. INTRODUCTION ............................................. 1
2 . PHILADELPHIA DATA BASE ................................... 3
Emissions .............. ............ .................. 5
Concentrations ..... ... ............. .... .............. 7
Meteorology .......................................... 11
3. MODEL EVALUATION ......................................... 17
PEM-2 Runs ........................................... 17
Calculation Grid ................................. 18
Emission Data .................................... 20
Deposition and Transformation Rates .............. 31
Model Calculations ............................... 32
Background Concentrations ....... . ....... . ........ 33
Evaluation Statistics ................................ 34
4. RESULTS AND DISCUSSION ................................... 39
Sulfur Dioxide ...... ......... . . ...................... 39
Sulfate ...................... . ....................... 55
Fine Total Mass ...................................... 65
Coarse Total Mass ..... ............................... 70
-------
FIGURES
Number Page
1. Philadelphia area map with locations of monitoring
stations in PAFS field program 8
2. Average daytime wind distribution during PAFS 13
3. Average nighttime wind distribution during PAFS 14
4. Frequencies (No. of hours) of PG stability classes
during PAFS 16
5. Calculation grid used for PEM-2 evaluation showing
the locations of PAFS stations , 19
6. Variations of daily total (daytime + nighttime) S02
emissions from point and area sources in Philadel-
phia for the evaluation period 21
7. Same as in Fig. 6, except for sulfate emissions 22
8. Same as in Fig. 6, except for fine total mass
emissions ..».. ....... 24
9. Same as in Fig. 6, except for coarse total mass
emissions 25
10. Diurnal variation of S02 emissions (averaged over
the evaluation period) from ooint and area sources
in Philadelphia 26
11. Same as in Fig. 10, except for sulfate emissions 27
12. Same as in Fig. 10, except for fine total mass
emissions ...,.,.> >..t......... 28
13. Same as in Fig. 10, except for coarse total mass
emissions ..<,..< < . .............. 29
14. Comparison of calculated and observed diurnal
variations of S02 concentrations (averaged over the
evaluation period) at Station 5 . c 40
15. Same as in Fig. 14, except at Station 7 41
16. Same as in Fig. 14, except at Station 8 .,, 42
17. Same as in Fig. 14, except at Station 12 43
18. Same as in Fig. 14, except at Station. 28 44
vi
-------
FIGURES (continued)
19. Same as in Fig. 14, except at Station 34 45
20. Comparison of calculated and observed 12-hour average
S(>2 concentrations for the evaluation period 48
21. SC>2 residuals (D^^ - 0^ - P^ ) versus observed 12-hour
average SC>2 concentrations for the evaluation period .. 50
22. Comparison of calculated and observed daily mean
concentrations of SC>2 (averaged over all PAFS
stations) for the evaluation period 52
23. Same as in Fig. 22, except for daytime (12-hour
average) S02 concentrations 53
24. Same as in Fig. 22, except for nighttime (12-hour
average) SC>2 concentrations 54
25. Comparison of calculated and observed 12-hour average
sulfate concentrations for the evaluation period 56
26. Sulfate residuals (D^ = 0^ - P^ ) versus observed
sulfate concentrations for the evaluation period 59
27. Comparison of calculated and observed daily mean
sulfate concentrations (averaged over all PAFS
stations) for the evaluation period 62
28. Same as in Fig. 27, except for daytime (12-hour
average) sulfate concentrations 63
29. Same as in Fig. 27, except for nighttime (12-hour
average) sulfate concentrations 64
30. Comparison of calculated and observed 12-hour average
FP total mass concentrations for the evaluation
period 66
31. FP total mass residuals (Dj_ = 0± - P^ ) versus
observed 12-hour average FP concentrations for the
evaluation period 68
32. Comparison of calculated and observed daily mean FP
total mass concentrations (averaged over all PAFS
stations) for the evaluation period 71
33. Same as in Fig. 32, except for daytime (12-hour
average) FP total mass concentrations 72
34. Same as in Fig. 32, except for nighttime (12-hour
average) FP total mass concentrations 73
vii
-------
FIGURES (continued)
35. Comparison of calculated and observed 12-hour average
CP total mass concentrations for the evaluation
period 74
36. CP total mass residuals (D^ =• Oj - PA ) versus
observed 12-hour average CP concentrations over the
evaluation period 76
37. Comparison of calculated and observed daily mean CP
total mass concentrations (averaged over all PAFS
stations) for the evaluation period 79
38. Same as in Fig. 37, except for daytime (12-hour
average) CP total mass concentrations „ 80
39. Same as in Fig. 37, except for nighttime (12-hour
average) CP total mass concentrations ................. 81
viii
-------
TABLES
Number
1. PEM-2 evaluation days of PAFS data .. 4
2. Average daytime and nighttime total emission rates from
area and point sources in PAFS inventory 30
3. Summary of PEM-2 evaluation statistics for S02 49
4. Summary of PEM-2 evaluation statistics for sulfate 57
5. Results of stepwise regression analysis for sulfate 61
6. Summary of PEM-2 evaluation statistics for fine
total mass 67
7. Results of stepwise regression analysis for fine
total mass 69
8. Summary of PEM-2 evaluation statistics for coarse
total mass 75
9. Results of stepwise regression analysis for coarse
total mass 78
10. Comparison of daily average total emission rates from
area and point sources in Philadelphia and St. Louis ...... 86
ix
-------
ACKNOWLEDGEMENTS
This report ts prepared for the Office of Research and Development,
Atmospheric Sciences Research Laboratory (ASRL) of the U. S. Environmental
Protection Agency (EPA) to support the needs of the Office of Air Quality
Planning and Standards in urban particulate modeling. This work is
accomplished under interagency agreements among the U.S. Department of
Energy, the National Oceanic and Atmospheric Administration (NOAA), and
the EPA.
The authors thank James Godowitch and Jack Shreffler of ASRL for the
guidance and advice during the course of this work, and for their interest
and patience. One of the authors (JYK) would like to express his
appreciation to the personnel of NOAA's Atmospheric Turbulence and
Diffusion Division (ATDD), especially to Rayford Hosker and Bruce Hicks for
arranging his visit to ATDD to participate in this project.
-------
SECTION 1
INTRODUCTION
The U. S. Environmental Protection Agency (EPA) recently proposed
revisions to the National Ambient Air Quality Standard (NAAQS) for
inhalable particulate (IP) matter which would base the primary,
health-related standard on particles smaller than 10 microns aerodynamic
diameter (PM-10). As part of the effort to control urban particulate
emissions, EPA sponsored the development and evaluation of an improved
urban scale model capable of simulating the transport, diffusion, and
deposition of particulate matter.
The Pollution Episodic Model Version 2 (PEM-2) described by Rao (1985)
is an urban scale (10-50 km) model capable of predicting short term (1-24
hr) ground-level concentrations (GLC) and deposition fluxes of one or two
gaseous or particulate reactive pollutants in an urban environment with
multiple point and area sources. It is intended primarily for particulate
modeling, and also for studies of the atmospheric transport, transfor-
mation, and deposition of acidic, toxic, and other pollutants to assess the
impact of existing or new sources or source modifications on urban air
quality. The PEM-2 concentration algorithms (Rao, 1983) explicitly account
for the effects of dry deposition, sedimentation, and a first-order
chemical transformation.
PEM-2 is based on the Pollution Episodic Model (PEM) developed by Rao
and Stevens (1983). The latter, in turn, is based on the Texas Episodic
-------
Model Version 8 (TEM-8)- developed by the Texas Air Control Board (1979) for
the atmospheric dispersion of non-reactive pollutants over a perfectly
reflecting (non-depositing) surface. Rao (1985) discussed the key
differences between these various models, and gave the algorithms,
computational techniques, capabilities, limitations, and input/ouput
parameters of PEM-2. Pendergrass and Rao (1984) described an evaluation of
PEM using the St. Louis Regional Air Pollution Study (RAPS) data.
This report describes an evaluation of PEM-2 using the data from EPA's
1982 Philadelphia Aerosol Field Study (PAFS). This evaluation is designed
to test the performance of the model by comparing its concentration esti-
mates to the measured air quality data, using appropriate statistical
measures of performance.
Twenty nine days (from July 16 to August 13, 1982) in the Philadelphia
data base are utilized for the PEM-2 evaluation. The model performance is
judged by comparing the calculated average concentrations with the
corresponding observed values for the following pollutant species:
1. S02 2. Sulfate particles
3. Fine total mass 4. Coarse total mass
The cut-point size between fine particle (FP) and coarse particle (CP)
total mass fractions is 2.5 microns* A first-order chemical transforma-
raation of SC>2 to fine sulfate is considered in the calculations in addition
to the direct emission and dry deposition of all four pollutant species.
For each pollutant, several plots examining the model performance are
given, and the model evaluation statistics are tabulated.
-------
SECTION 2
PHILADELPHIA DATA BASE
The 1982 Philadelphia Aerosol Field Study has been sponsored by the EPA
to obtain detailed data bases on emissions and surface air concentrations
of gases and particles, and concurrent meteorological conditions, with
adequate temporal and spatial resolution to evaluate source and receptor
models, including PEM-2, which are capable of predicting short term
concentrations of particles. The emissions of primary particulate matter,
and pollutants that undergo chemical transformation in the atmosphere to
form secondary particulates are of special interest. Therefore, the study
was designed to obtain data on total particulate mass, sulfate, and SC>2.
Philadelphia was chosen for this study because it has a good mix of
industrial emissions, and it was the site of an earlier program for
evaluation of EPA's Urban Airshed photochemical oxidant model.
The PAFS experiment was conducted from July 14 to August 14, 1982.
From this data base, detailed emission inventories, and meteorology and
concentration measurements for twenty nine days are supplied by the EPA for
PEM-2 evaluation. The evaluation days are listed in Table 1. The PAFS
experiment and data base have been described in detail in other publica-
tions referred to below, and will not be discussed here; only the data used
in the evaluation of PEM-2 and other relevant information will be described
in this section.
-------
TABLE 1
PEM-2 Evaluation Days of PAFS Data.
Pay Date Julian Day
1 July 16, 1982 197
2 17 198
3 18 199
4 19 200
5 20 201
6 21 202
7 22 203
8 23 204
9 24 205
10 25 206
11 26 207
12 27 208
13 28 209
14 29 210
15 30 211
16 31 212
17 August 1, 1982 213
18 2 214
19 3 215
20 4 216
21 5 217
22 6 218
23 7 219
24 8 220
25 9 221
26 10 222
27 11 223
28 12 224
29 13 225
-------
EMISSIONS
Hourly emission inventories for point and area sources in the
Metropolitan Philadelphia area during the PAFS monitoring period are deve-
loped by Engineering-Science (ES) of Fairfax, Virginia. The inventories
are listed for FP total mass (diameter < 2.5 microns), CP total mass (2.5 -
10 microns), primary sulfate particles, and sulfur dioxide. The primary
sulfate data are found from a variety of industrial processes, and its
emission factors are expressed as multipliers of S02 values on a weight
basis or as percent of flyash. These data are for total sulfates which
include the sulfuric acid aerosol. No data are found on sulfate particle
sizes. Because these are thought to be in the lower end of the IP size
range, ES (1984) assumed that all sulfate particles are less than 2.5
microns in diameter. Therefore, emission data on coarse primary sulfate
particles are not available, and coarse sulfate emissions are assumed to be
zero in this evaluation.
The emission inventories are supplied on two magnetic tapes, the first
with 300 major point sources, and the second with 289 area sources; the
latter are given on a square grid with 17 x 17 cells each of which is a
square of side 2.5 km. This emission grid covers the Philadelphia county
and portions of Bucks, Chester, Delaware, and Montgomery Counties in
Pennsylvania, and Burlington, Camden, and Gloucester Counties in New
Jersey. Though this area is smaller than the Air Quality Control Region
(AQCR), the grid encompasses the Philadelphia urban area including all of
Philadelphia County to the North and East, and extends well beyond the
monitoring site locations to the West and South, thus retaining all the
emission data important for model evaluation.
-------
In addition, ES provided emission data for minor point sources and
fugitive particulate matter on a third tape, and for mobile sources on a
fourth tape. These emission data are given on a square grid with 17 x 17
cells each of which is a square of 2.5 km side. This grid coincides with
the area source grid, so that these emissions can be easily merged with
area sources in the model calculations. Highway vehicle emissions data are
generated by the Delaware Valley Regional Planning Commission (DVRPC) under
subcontract to ES. The latter conducted limited source testing and
collected silt loading samples from selected paved roads to provide
site-specific data in Philadelphia for model evaluation.
The point source emission data contained the stack identification, its
location in UTM (x,y) coordinates, stack parameters such as height,
diameter, plume-exit velocity and temperature, and emission rates of the
four pollutants (S02, sulfate, fine mass, and coarse mass) in g/s. The
area source emission data included the county code, grid cell number,
location of southwest corner of the source in UTM (x,y) coordinates, and
emission rates (in g/s) of the four species from each grid cell.
The major point sources in Philadelphia are electric utilities using
oil and coal-fired boilers , municipal incinerators, oil refineries, metal
and chemical industries, grain elevators, etc. The stack heights are
generally below 100 m. Outside a 42.5 km x 42.5 km inventory area, ES
selected only about 50 (out of 300) major point sources with SOX or total
mass emissions above a cutoff value of 500 tons per year. All other point
sources with relatively lower emissions or farther away from the monitoring
sites are classified as minor point sources. The area sources consist of
residential, commercial, industrial, and agricultural emissions, and
-------
releases from transportation, incineration, construction, and other
activities. ES developed procedures to generate hourly emission rates
using data they collected within the framework of the PAFS study. For a
discussion of these procedures and additional details of the point and area
source inventories, the emissions report by Engineering-Science (1984)
should be consulted.
It should be noted that these emission data do not represent real time
conditions. For example, the precipitation events are not taken into
account in developing the hourly emission estimates. The use of relatively
large (2.5 km square) grid cells in the study complicates model evaluation
since impacts from paved roads very near the monitors are likely to be the
most significant contributions. Model results should be interpreted keeping
these limitations in mind.
CONCENTRATIONS
The ambient air monitoring network in the PAFS study was operated by
PEDCO Environmental, Inc. of Cincinnati, Ohio. The actual period of data
gathering lasted from 6 a.m. Eastern Daylight Time (EOT), July 14 to 6 a.m.,
August 14, 1982. The locations and characteristics of the six monitoring
sites shown in Fig. 1 are described below:
Site 5
Philadelphia, PA
Community Health Services Building
500 South Broad Street
Center City - Commercial
-------
_ o
-------
Site 7
Philadelphia, PA
Fire Boat Station
Alleghany Ave. & Delaware River
Suburban - Industrial
Site 8
Philadelphia, PA
Water Treatment Plant
Ford Road & Belmont Ave.
Suburban - Residential
Site 12
Philadelphia, PA
Philadelphia Northeast Airport
Grant and Ashton Road
Suburban - Rural
Site 28
Camden, NJ
Institute for Medical Research
Copewood and Davis Streets
Suburban - Residential
Site 34
Clarksboro, NJ
Shady Lane Home
Cohawkin Road and County House Road
Suburban - Rural
The monitors at Site 5 are located on top of a building (11 m above
street level) and are shielded to the south by a 2.5 m high building
extension. Therefore, these monitors probably are not collecting
representative samples from either the street-level or the building-top
level. Site 7 is the most impacted by fugitive dust due to nearby truck
terminals and heavy traffic on an adjacent paved road. The exact location
of the roadway and construction-related emissions are not input to the
model. Site 8 is affected by particulate emissions from a county
maintenance yard with unpaved roads and aggregate storage piles. Site 12,
-------
located near an area used for plane parking, is affected by heavy traffic
on roads in the vicinity as well as airport operations and nearby small
point sources. Site 28 is expected to be significantly impacted by paved
road emissions, and Site 34 by a number of large point sources in the
vicinity.
The measurements at the six sites listed above consist of the
following:
1. Continuous gas monitoring of sulfur dioxide (S02), nitric oxide (NO),
nitrogen dioxide (N02), non-methane organic carbon (NMOC), ozone (03),
and carbon monoxide (CO), to obtain 1-hour average concentrations.
2. Particulate mass sampling twice daily at each site to obtain two
12-hour average concentrations, one for daytime (6 a.m. to 6 p.m.)
and the other for nighttime (6 p.m. to 6 a.m.).
3. Meteorological data - wind speed, wind direction, doppler sodar data
(at Site 28 until August 5, then relocated at Site 8), and minisondes
(at Site 28 only).
These measurements are designed to obtain sufficient data to model FP
and CP total mass, and sulfate concentrations on a 12- and 24-hour basis.
S02 is monitored hourly to determine the chemical transformation
contribution to the sulfate concentration. The other gases, namely, NO,
N02, NMOC, and 03 are monitored to establish sulfate mass formation via
photochemical mechanisms. The study is conducted 24 hours a day to obtain
data for modeling diurnal variations in sulfate formation. Sites 28 and 34
are operated by the State of New Jersey, and the other four sites by the
City of Philadelphia. PEDCO installed and operated additional monitoring
equipment for PAFS to supplement the instrumentation available at each
site.
10
-------
The continuous gas analyzers for SC>2 are either Beckman 953 or TECO
Series 43; the particulate monitors are EPA's high-volume or dichotomous
filter samplers. For the duration of PAFS, PEDCO operated the various
monitors, recorders, and data acquisition systems, and performed the
necessary checks, audits, and calibrations called for in the Quality
Assurance (QA) Project Plan for the field study. For details on these
procedures and instrumentation used in PAFS, the reader is referred to the
report by PEDCO (1983).
For PEM-2 evaluation, the measured concentrations of SC>2, fine and
coarse sulfate particles, and FP and CP total mass from the six monitoring
sites of PAFS are provided by EPA on a magnetic tape. All concentrations,
except S02, are 12-hour averages; S02 data are 1-hour averages.
METEOROLOGY
Each of the monitoring stations is equipped with continuous wind speed
and direction instruments. Data from each instrument are collected on
strip chart recorders and data acquisition system. Aerovironment, Inc.
(AV) performed thrice daily (4 a.m., 10 a.m., 4 p.m.) soundings of pressure
(height), temperature, and wet-bulb depression (relative humidity) at
Site 28 using an airsonde system developed by the Atmospheric
Instrumentation Research (A.I.R.). Upper air wind data are obtained by
using a theodolite to track the balloon that lifts the airsonde. To
monitor upper air meteorology, a monostatic Doppler acoustic sounder is
operated by AV to provide 15 minute averages of horizontal wind speed and
direction at 30 m height increments up to 1 km.
11
-------
The wind speed and wind direction data recording systems malfunctioned
during the PAFS experiment; the surface meteorological measurements at the
six sites are either missing or appear to be of dubious quality. There-
fore, EPA has decided to use the hourly surface winds and temperatures
observed at the Philadelphia International Airport (PHL) National Weather
Service (NWS) Station (see Fig. 1) as input data for PEM-2 evaluation. A
magnetic tape of these data consisting of hourly values of wind speed and
direction, temperature, stability class, and mixing heights has been
provided by the EPA. For each hour, the Pasquill-Gifford (PG) stability
class and the two sets of mixing heights are determined by the RAM
preprocessor (RAMMET), following a procedure described by Turner and Novak
(1978). Morning and afternoon mixing heights required by RAMMET are
derived from the observed airsonde temperature profiles as the base of the
elevated inversion.
The values of the potential-temperature gradient above the mixing
height are also determined from the temperature profiles for use in the
optional new plume-penetration schemes of PEM-2 (see, Rao, 1985). The
meteorological data are compiled into data files suitable for input to
PEM-2; for example, stability class 7 (PG class G) in the RAMMET output is
reset as stability class 6 (PG class F) in FEM-2.
Figure 2 shows the daytime wind distribution from the NWS data at PHL
during the PAFS experiment. The surface winds during the day are
predominantly from the southwest (WSW to SSW), with speeds ranging mostly
from 2 to 6 m/s. The nighttime (6 p.m. to 6 a.m.) wind rose presented in
Fig. 3 also shows a dominant southwesterly component. The major point
sources in the Philadelphia area are distributed along a line oriented
12
-------
TJAY
NNE
ENE
Figure 2. Average daytime wind distribution during PAFS.
13
-------
"NTURT
NNE
Figure 3. Average nighttime wind distribution during PAFS.
14
-------
approximately from SW to NE. Hence, the PAFS station receptors should
receive significant contributions from point sources both day and night.
The nighttime wind rose also shows a strong northerly component; the
nocturnal winds are evenly distributed from WSW to NW.
The frequencies of the six PG stability classes in the hourly
meteorological data used in model evaluation are shown in Fig. 4. The
slightly unstable (class C) cases during the day, and moderately stable
(class F) cases during the night, occur frequently. There are roughly
equal number of daytime and nighttime neutral cases. The frequencies of
class B (moderately unstable) during day and class E (slightly stable)
during night are also about equal.
15
-------
I
I
UJ
Q C/)
(ft
o
CD
<
CD
O O
O 03
M
o
2
co en
-------
SECTION 3
MODEL EVALUATION
The details of the PEM-2 computer runs, input parameters, and
statistical procedures of the model evaluation are discussed in this
section.
PEM-2 RUNS
The PEM-2 concentration predictions are evaluated against the measured
concentrations of four pollutant species:
1) S02
2) Fine sulfate particles
3) FP total mass
4) CP total mass
The coarse sulfate concentrations could not be evaluated separately since
coarse sulfate emissions are assumed to be zero in the PAFS inventory (see
Section 2). The four species are calculated in two model runs as follows:
Run Pollutant-1 Pollutant-2 Note
I S02 Fine sulfate Chemical transformation
of S02 to fine sulfate
is considered.
II Fine mass Coarse mass No chemical coupling
between the species.
The fine total mass emissions in Run II also include the primary sulfate
17
-------
particles. Though the emissions of the latter are assumed to consist only
of fine particles, they obviously include some coarse (2.5 - 10 microns)
sulfate particles as well since, for some point sources in the PAFS
inventory, the primary sulfate emission rates exceed the corresponding fine
total mass emission rates. Furthermore, the chemical transformation (S02
to fine sulfate) contribution to the calculated fine total mass
concentration is ignored in Run II. These errors, however, are not
expected to be serious for obtaining the urban-scale 12-hour average
concentration estimates based on the entire inventory in this evaluation.
Calculation Grid
The calculation grid consists of 32 x 32 cells, each a square of 2.5 km
side. The southwest corner of the grid is set at XUTM « 444.5 km and YUTM
= 4380 km, and the modeling domain covers 80 km x 80 km to incorporate the
entire PAFS emission data. The calculation grid cell size is chosen to
agree with the area source grid cell size of 2.5 km, so that emissions can
be directly input to the model from the inventory data files. Figure 5
shows the calculation grid used for PEM-2 evaluation and the locations of
the six PAFS stations.
For this evaluation, the maximum capacity of 50 area sources in PEM-2
is increased to accomodate all 289 area sources in the emission Inventory.
The PEM-2 program is also modified such that concentrations from point
sources are calculated only at the four receptors (grid points) surrounding
each PAFS station, and not at the rest of the receptors. This required
calculation at only 24 out of a total of 1024 receptors, which resulted in
a significant reduction in the computer run costs. The area source
18
-------
o
U3-
O
CJ
,-|
-V2-
a
o
01:
28
TTT
H-.I-T"!
I
1.5
Figure 5.
l.S
484.5
XUTM (km)
S04.S
524. S
Calculation grid used for PEM-2 evaluation showing
the locations of PAFS stations.
19
-------
calculations did not include this modification to the program. For each of
the area sources, the contributions to the concentrations at the nine
affected receptors located within and immediately downwind of the source
are calculated, as discussed by Rao (1985). The total concentration at
each PAFS station is then determined as the weighted average of the
corresponding values calculated at the four surrounding receptors.
A. calculation grid cell size of 5 km side square is also tested in the
evaluation. The results showed that, for the PAFS emissions inventory, the
calculations are not very sensitive to this change in the grid cell size.
Emission Data
The hourly emission data are scanned and point sources with emissions
less than 1 g/s of S02 and 0.1 g/s of 804 are eliminated in order to reduce
computer run costs. The emissions are analyzed for day-to-day as well as
diurnal variability.
Figure 6 shows the variations of daily total (daytime + nighttime) S02
emissions from all of the point and area sources in the PAFS inventory
during the evaluation period. Figure 7 shows a similar plot for daily
total sulfate emissions. It can be seen that the point-source total
emissions show large day-to-day variability; for example, S(>2 and sulfate
emissions increase by roughly 50 percent from Day 17 to Day 18, and
decrease by a similar amount from Day 22 to Day 23. Days 18 to 22 (August
2-6) are characterized by unusually large point-source emissions, most
probably due to increased power demand for air-conditioning during the hot
summer days. In contrast, the area-source total S02 and sulfate emissions
20
-------
o
O
00
UJ
CJ
O
00
o
a.
a
01 -O
CN
OUT:
en .£ a>
4-1 4-1
o
*-> tJ C
» « T3
rH (U C
•H JJ 3
C8 fl «
T3
•O 40
«-i e w
o as ^
•H
CO 4J fcj
C C
O i-l
•H O .
4J a. TJ
« o
T-l E 1-1
l-i O U
CO l-i MH a.
vO
0)
I-l
M
(s/6>0 SNOISSIW3
21
-------
UJ
O
o:
ID
O
00
<
LU
UJ
O
a:
ID
O
00
O
a.
c/)
<
i
<
i
<
4
f
<
i
4»
(N
o
(N
CO
o
e
0)
0)
4J
(0
3
03
la
O
4J
a.
a)
o
X
- vO
m
-------
are constant during weekdays, and decrease during weekends to their minimum
values on Sunday.
Figures 8 and 9 show variations of daily total (daytime + nighttime)
emissions of FP and CP total mass, respectively. The point-source fine
mass emissions show significant day-to-day variability during the
evaluation period, while the coarse mass emissions do not vary much. The
area source emissions of both fine and coarse mass show periodic
variations, typically increasing slightly during weekdays to a maximum
value on Friday, and decreasing during weekends to a minimum on Sunday.
These variations are probably dictated by traffic patterns and fugitive
particulate emissions from unpaved highways, and emissions from industrial
and construction activities.
The diurnal variations of SC>2 and sulfate emissions from point and area
sources in Philadelphia (averaged over the 29-day evaluation period) are
shown in Figs. 10 and 11, respectively. The point-source emissions
increase sharply after 7 a.m. to their daytime peak values and then
decrease rapidly after 8 p.m. The area-source emissions also behave in a
similar manner, though these values are much smaller than the point-source
emissions. However, area sources dominate the daytime emissions of fine
and coarse total mass as shown in Figs. 12 and 13, respectively. The total
particulate emissions increase sharply from 5 a.m. to a daytime peak at
7 a.m., and decrease rapidly from a secondary peak at 4 p.m. Thus, these
emissions strongly correlate with rush-hour traffic patterns and daytime
industrial activity.
23
-------
UJ
o
ID
O
CO
!4i LO
05 CO
uj O
O i-
CC Ul
=3 Z
o cz
CO
O
Q_
•q
If)
CN
O
IN
m •<
"• Q
CD
e
o
T-l
to
en
39
03
ffl
as
j_i
o
|) SNOISSIW3
24
-------
UJ
o
(XL
< in
uj £
cr <
< 3
o
C/)
O
Q_
O
<
HJ--
in
CN
o
CN
in <;
*" Q
<£>
cs(
CO
c
o
•H
CO
OJ
•H
1) SNOISSW3
25
-------
CJ
o
CO
LJ
o
ce
z>
O
00
O
Q_
» w
D
•H C
s -t
0) O
O.
tM
O S
W O
t-t
U-l o o.
•H r-l
i— i ij ai
« n -a
C 3 *
td i-l r-l
3 03 i-l
•H > j:
a
-------
i r
O
cvl
01
C
o
•H
co
CO
UJ
O
a:
ID
O
oo
0)
u
(0
u-i
i— I
3
tn
(8
3
t>0
T-l
o
o
SNOISSIIAG
27
-------
Ld
O
33
O
,-j-
fe! °°
C£ oo
"^
< *
j<
Ld O
o i-
Qi 1 1 1
N -—*
1 Z
o u:
w /
1 —
21
O
Q_
» c
D
i , , . , , ^ . ^j, ,
m 3 I
X
- /^
]
1
]
,
"
-
.
-
3
X \
™* "*" ~" *" "" — — — -». ^^\.
'~"~"**"~'*'™'*-«3 ^X
-------
UJ
O
o:
ID
O
00
UJ
ce
UJ
o
ce
z>
o
00
00
00
R
UJ
00
cc
<
O
o
1- 0 SNOISSIW3
29
-------
The total emission rates of the four pollutant species from point and
area sources, averaged over the evaluation period and stratified by day and
night, are shown in Table 2. It can be seen that point sources dominate
the SC>2 and sulfate emissions. The area sources dominate the fine and
coarse mass emissions during the day; however, the contributions of point
and area sources are roughly equal at night.
TABLE 2
Average Daytime and Nighttime Total Emission Rates (kg/s) from
Area and Point Sources in PAFS Inventory.
Pollutant
Area Sources
Point Sources
S02
Sulfate
Fine Mass
Coarse Mass
Daytime (6
Nighttime
:00 a.m. - 6:00 p.m.)
16.665
0.650
7.897
5.133
(6:00 p.m. - 6:00 a.m.)
14.683
0.555
3.021
1.668
56.425
1.797
4.269
1.737
46.697
1.480
3.724
1.536
S02
Sulfate
Fine Mass
Coarse Mass
Comparing this with a similar tabulation of total emissions for the RAPS
study from Pendergrass and Rao (1984), we note that unlike in St. Louis,
30
-------
area sources in Philadelphia are significant contributors of SC>2 and
sulfate, and point-source contributions to fine and coarse total mass are
very important.
Deposition and Sedimentation Rates
Rao (1983) discussed the specification of deposition and gravitational
settling velocities (V
-------
during day. The sulfate formation rates are smaller at night due to the
absence of photochemical mechanisms. A constant transformation rate of 3
percent per hour is used for the nighttime runs in this study.
Model Calculations
The PEM-2 is run with hourly data of meteorology and emissions to
estimate the hourly concentrations. The output data are stored on a
magnetic tape, and later summed and averaged to obtain 12-hour average (day
and night) values. The model evaluation runs are made with the following
technical options (see, Rao, 1985) in effect:
1) Wind-profile exponents option: NWPOPT = 0, standard values are used;
2) Stack-tip downwash option: NSTDWN = 0, downwash algorithm is used;
3) New plume-rise equations option: NPRISE = 1, new equations are used
to estimate maximum rise of buoyancy-dominated plumes from point
sources in unstable or neutral atmosphere;
4) New plume-penetration schemes option: NINPEN = 1, new schemes are used
for estimating penetration of an elevated stable layer (capping
unstable/neutral atmosphere) by buoyancy-dominated plumes from
point sources.
The default values of atmospheric potential temperature gradients for E
and F stability classes (0.02 and 0.035°C/m, respectively) are used in the
plume rise equations under stable conditions. The area source emissions
are assumed to be located at an effective height HAS = 10 m, which is a
typical building height in urban areas. This value is kept constant for
all area sources and species both day and night.
32
-------
Background Concentrations
The calculated 12-hour average sulfate, fine and coarse total mass con-
centrations, resulting only from the contributions of point and area sour-
ces to the receptors, are added to their respective background
concentrations. The latter are determined as the lowest observed 12-hour
average concentrations at one of the four PAFS stations 8, 12, 28, and 34
located in a generally upwind direction during the averaging period.
Stations 5 and 7, which are located in the downtown urban area and have
higher observed concentrations compared to the other stations, are not used
for this background analysis. The background concentrations of S02 are
assumed to be zero.
The method of determining the background concentrations as described
above is rather subjective due to the small number of monitoring stations
available. However, the background particulate concentrations play an
important role in evaluating the model performance by comparing the
calculated concentrations to their observed values. For example, the
calculated fine mass concentrations directly attributable to the emissions
in Philadelphia are in the range of 0 - 20 Ug/in3 with an average of about
5 Vig/m3, while the background fine mass concentrations average about
25 Ug/m3. The observed coarse mass concentrations at the stations are more
variable than for the fine mass; this suggests that local contributions may
be more important for coarse mass. The calculated average concentration of
coarse mass resulting from local emissions is below 5 Ug/m3, while its
background concentration is about 10 Ug/m^.
The station-to-station variations of the observed sulfate concentra-
tions are small, with the background values averaging about 7
33
-------
On the other hand, the sulfate concentration calculated from direct
emissions and chemical transformation is below 2 yg/m^. Therefore, the
background concentrations of particulate species play an important role in
this evaluation.
EVALUATION STATISTICS
The recommendations of two workshops sponsored by the American
Meteorological Society (AMS) to review the statistical approach to air
quality model evaluation and model uncertainty are summarized by Fox (1981,
1984). Discussions and applications of these statistical methods can also
be found elsewhere in the literature (see, e.g., Ruff, 1983; Rao et al.,
1985).
The predicted and the corresponding observed concentrations are treated
as pairs in this evaluation. Two general measures of performance are used
here: (a) measures of difference which include the bias, variance, gross
variability or root mean squared error, average absolute gross error, mean
fractional error, and index of agreement; (b) measures of correlation
including the correlation coefficient, slope, and intercept; estimates of
the predicted concentrations from the regression analysis allow determin-
ation of the systematic and unsystematic parts of the mean squared error.
The observed and predicted concentrations are analyzed and plotted with a
standard SAS statistical and data-handling package (Ray, 1982).
In the discussion that follows, the observed concentrations are denoted
by 0± and the corresponding predicted concentrations (paired in space and
time) are denoted by P±< A11 sums are caicuiated over i = 1,2, ,N,
34
-------
where N is the number of observations. The means, 0 and P, and standard
deviations (So and Sp) are computed as
where OJ - Q± - 0" I 0^ = 0 (3)
and P - P - 7 , I P = 0 (4)
±
The mean and standard deviation of the ratios, PI/O^ , between the
predicted and its associated observed value are also computed.
(a) Measures of Difference
Differences (residuals) are based on the observed and calculated
concentrations such that
DI - Oi - PI (5)
A negative residual indicates model overprediction and vice versa. The
bias D, which is the first moment of the distribution of differences, is
defined as
D-TT-P-IXDI (6)
This is a measure of the overall bias of the model in predicting pollutant
concentrations.
35
-------
The average absolute gross error is defined as
This measure of the absolute size of the error is less affected by the
removal of outliers in the data than the root mean square error
(RMSE). The estimated variance, which is the second moment of the distri-
bution of differences, is calculated from
(8)
D! =• D. — D , l D! = 0
where S^ is the standard deviation. The variance is a measure of noise in
the data. The RMSE, which is a measure of the actual size of the error
produced by the model, is computed from
2
BMSE
[(TT)
The mean fractional error (MFE), which determines the model's overall bias
to underpredict or overpredict the concentrations, Is given by
MFE = I I i :rT7T (10)
°i + Pi)/2
']
For normally distributed variables, the bias has a normal distribution
2
while the variance S, has a chi-squared distribution. The mean square
error has a compound distribution. If the distributions of the predicted
and observed concentrations are the same, then it is reasonable to assume
that the differences are normally distributed with a zero mean and a
constant variance.
36
-------
The index of agreement is a measure of the degree to which the observed
variable is accurately estimated by the calculated variable. This is not a
measure of correlation, but rather a measure of the degree to which the
model predictions are error-free. At the same time, it is a standardized
measure so that cross-comparison of its magnitude for different models, or
a model's performance at different receptor locations or under different
atmospheric conditions, can be made. The index of agreement, d, is
expressed as
Thus, d specifies the degree to which the observed deviations about 0
correspond, both in magnitude and sign, to the predicted deviations from 0.
It is assumed that the parts of the magnitudes of P± and 0± that are
equivalent to 0 are not in error since 0 is considered to be error-free.
All the potential for error is therefore assumed to be contained in the
deviations of P^ and 0^ from 0.
(b) Measures of Correlation
For each pollutant, a scattergram of predicted (on the ordinate)
versus observed (on the abscissa) concentrations is plotted, and linear
regression analysis is performed to determine a correlation coefficient
(R), slope (b), and intercept (a), as follows:
I (OJ • Pp
37
-------
I (0! • P!)
b = == — , a = P - b 0 (13)
The estimate of the predicted concentration, P^, is then given by
Pi - a + b 0± (14)
The unsystematic and systematic parts of the mean squared error (MSB)
are computed as follows:
MSE(u) -i I (P± - Pi)2 (15)
MSE(s) - i I (P^ - Ot)2 (16)
These new measures illuminating the sources or types of errors can be
helpful in refining a model. When the MSB is largely systematic, further
refinement in the model may be necessary in order to minimize the MSB so
that the model can predict at its maximum possible accuracy. On the other
hand, if MSE is largely unsystematic, the model is probably as good as it
can be, and may not require major modifications. For each species, the
complementary ratios MSE(u)/MSE and MSE(s)/MSE are computed and expressed
as percentages.
38
-------
SECTION 4
RESULTS AND DISCUSSION
The PEM-2 is evaluated using 29 days of data from Philadelphia for four
pollutant species: S02> sulfate, fine and coarse mass. Only S02 concentra-
tions are measured hourly; the particulate data are 12-hour averages. The
evaluation results comparing the model's concentration estimates to the
corresponding observed values are presented and discussed in this section.
SULFUR DIOXIDE
The hourly S02 concentration data provide a demanding test of short
term models such as PEM-2, especially since the background S02 concentra-
tions are assumed to be zero. Figure 14 shows a comparison of the calcu-
lated and observed diurnal variations of S02 concentrations (averaged over
the evaluation period) at Station 5. Similar plots for the other moni-
toring stations are shown in Figs. 15 to 19.
At all stations, the calculated and observed concentrations are closer
during daytime than at night. The model generally overpredicts the hourly
average concentrations at night.
A significant overestimation of S02 concentrations by the model at
Stations 5 and 7, and also 28 to a somewhat lesser extent, is evident when
compared with the evaluation results at other stations. The overpredic-
39
-------
Lul
C/5
GD
O
O
UJ
g
z>
O
b
CO
( iu/67/) NOLLV&LN30NOO
40
-------
Q
LJ
>
cr
LJ
01
00
o
<
_l
ID
CJ
D
rx
2:
O
CO
^
I
^3
<
\
o
KO
o
CN
O
-------
Ld
GO
CD
O
Q
Ud
o
a
03
GO
O
IN
00
e
o
CO
ij
a
Ld
-------
UJ
>
cc
UJ
CO
m
O
< 2
Q
UJ
o
o
* •
a
00
e
o
(Q
u
OT
a
-------
Q
UJ
>
a:
UJ
CO
QD
O
UJ
§
H>
CJ
• •
a
CO
NOIlVdiN30NOO
44
-------
Q
UJ
UJ
CO
GO
o
CJ
a
NOIiVdIN30NOO
45
-------
tions at these stations during nighttime are more severe than during
daytime. The emission sources in Philadelphia area are located roughly
along an axis from southwest corner to northeast corner of the full
calculation grid. Stations 5, 7, and 28 are located in the region of major
area source emissions, and surrounded by many point sources. On the other
hand, Stations 8, 12, and 34 are located away from the region of high area
source emissions and may be impacted mostly by the point sources. At these
stations, the model underestimates the observed concentrations during day,
and overestimates during night.
From the above, it is clear that the model overprediction results
mostly due to errors in estimating concentrations from the urban area
sources using the narrow plume hypothesis and other simplifying assumptions
(see, Rao, 1983). Uncertainties in the area-source emission conditions
such as source height, plume rise, and building wake effects also
contribute to these errors. In general, the isolated point source emission
inventories are better defined and documented than those of the distributed
urban area sources. The pattern of urban model estimates being highest
under stable (nighttime) conditions and lower under unstable (daytime)
conditions is attributed by Turner and Irwin (1985) to the errors in
modeling low—level sources.
The plots show that the observed concentration reaches a maximum about
2 to 3 hours after the morning transition when the stratification changes
from stable to neutral or unstable. The fumigation process which brings
down the elevated plume from above the surface-based inversion may partly
account for the peaks observed soon after sunrise (Rao et al., 1985).
Though PEM-2 does not treat the fumigation process, the model simulates
46
-------
these peaks quite well; this suggests that the observed peak concentrations
may be related to the increases in SC>2 emissions in the morning (shown in
Fig. 10).
Figure 20 shows a comparison of the calculated and observed 12-hour
average SC>2 concentrations for the 29 evaluation days. This scatterplot is
a composite of all (both day and night) paired comparisons at the six PAFS
stations. The compared range of concentrations extends from 7.8 Wg/m ,
which approximates the instrument accuracy, to a cutoff value of 57 Vg/ra^
corresponding to 2.5 times the standard deviation, Sp. A linear
regression fit, with a slope of 0.33 and an intercept of 17, is also shown
in the figure.
Table 3 summarizes the model evaluation statistics for S02 for the
total, day, and night paired data comparisons. The mean and standard
deviation of the ratios P±/0± are 1.51 and 1.66, respectively. Of
particular interest is the better performance of the model during the day
than at night. The index of agreement (0.43 for daytime and 0.21 for
nighttime) suggests that PEM-2 calculations of S02 concentrations are more
accurate during the daytime than at night.
The differences D^ between observed and predicted concentrations
are plotted in Fig. 21 against the observed concentrations. There is a
tendency for the model to overpredict observed concentrations less than
25 Ug/m3. The bias D of -5.71 Ug/m3 indicates that PEM-2 overpredicts
S02 concentrations on the average. This overprediction is mostly due to
the nighttime cases for which the bias is -12.63 Ug/m^; the daytime S02
concentrations are slightly underpredicted. Because of the smaller
diffusivities and mixing depths, the nighttime GLC may result mostly from
47
-------
40
50
C 10
*
I
f»
I*
u
L
A 30
T
c
20
10
S02 CONCENTRATION ( )iq/s3 )
LtSEND: A = ! OSS, 8 = 2 3SS, ETC.
A
AA
A A A S 4 Aft A
A A
Aft A
AA A 3A C AA A A
3 AAAA A ; 4 A
3 9 AAAAA AA A A A
3A A AA A A A
A 3AH38M A AS A AA A A A
C AA BAASAAA A C A AA A A
4A A A A AS
ASAAA AAB A A A A A
A A38 AAAA A A A A A A
ASA A CCAD A A A A A
D AA A AA A AA 3 A AA
A A A A
20
30
OBSERVED
40
Figure 20. Comparison -of calculated and observed 12-hour average
SC>2 concentrations for the evaluation period.
48
-------
TABLE 3
Summary of PEM-2 Evaluation Statistics for
Observations
Range
Mean
S.D.
R
Slope
Intercept
Mean of (Pi/Oi)
S.D. of (Pi/Oi)
Bias
S.D. of Difference
Average Absolute Gross Error
RMSE
Index of Agreement
Mean Fractional Error
MSE(u)
MSE(s)
MSE(u)/MSE
MSE(s)/MSE
Note: The units of Range, Mean,
Total
Obs. Calc.
289
8-46 0-150
16.77 22.48
7.28 22.96
0.10
0.33
17.00
1.51
1.66
-5.71
23.36
15.42
24.00
0.27
-0.46
519.67
56.48
90%
10%
S.D. , Interci
Day
Obs. Calc.
147
8-46 0-91
17.02 16.03
7.85 15.71
0.18
0.35
10.02
1.07
1.46
0.99
16.28
11.45
16.25
0.43
0.75
237.55
26.53
81%
19%
apt, Bias, S.D
Night
Obs. Calc.
142
8-39 0-150
16.52 29.15
6.65 27.07
0.19
0.37
23.12
1.96
1.98
-12.63
27.29
19.53
29.99
0.21
-1.71
721.90
177.31
80%
20%
. of
49
-------
302 CONCENTRATION i |is/s3 i
LESEND: 4 = 1 3BS, 3 = 2 G2S. £TC.
50 *•
40
20
ft 10
E
S
i a
D
u
A -iO
L
-20
-30
A
A A A
A A
A HHfl AA
A A A A A A
H 3 AfiAft A A A
A A AAA AC AAA A A A
3 AB53EAAAA A A Afi A
EBSACSfl ABCA AAC A 3 A A
SCAAA 3AACA AA3 36 S AAS
-AASSAACSS-AftAA-G-iiA A
3CA3 3 5BAA 34 AA A
A 4SAA8 A 33 A A
AA A A A A A AA A A
A AAA A A A
A A A BAA A A A
A 3 3 A 3
ABAA A A A A
A AA A A AA A
A AA A A A A
A 8 A
3 A A A
-40
A A
-50
A
A
:o :o 40 so
OBSERVED CCNCEHTRATIOS
Figure 21. S(>2 residuals (D^ » 0^_ - P^) versus observed 12-hour
average 302 concentrations for the evaluation period.
50
-------
the surface area source emissions; on the other hand, point sources
contribute significantly during the daytime. If more of the emissions were
treated as point sources, the result would be to increase the daytime
concentrations, and decrease the concentrations at night. Thus, the model
results are sensitive to the characterization of emissions. Ruff (1983),
and Turner and Irwin (1985), identified the inclusion of point sources in
the RAPS emissions inventory as area sources as a possible cause of the
stability bias in the urban model (RAM) performance, whereby the largest
concentrations are often predicted to occur with stable conditions and low
wind speeds, in contrast to the observations.
Figure 22 shows a comparison of the calculated and observed daily
(24 hour average) concentrations of S02 (averaged over all six stations
for each of the 29 evaluation days). In general, the calculated daily
concentrations are within a factor of two of the corresponding observed
values, except for days 2, 18, and 23. If we plot the daytime and night-
time 12-hour average data separately (Figs. 23 and 24), we find that the
large differences between the calculated and observed concentrations
occured mostly during the night. These results confirm that the model
performs much better during the day than at night, as noted above. Even at
night, however, the major part of the error (about 80 percent) for SC>2 is
unsystematic. This means that, for the given data set, PEM-2 is predicting
with a high degree of accuracy, and we cannot readily identify where
further improvements can be made to the model.
In order to test whether the optional new plume rise equations and
plume penetration schemes of PEM-2 used in this evaluation made any
difference, the model is rerun with the standard plume rise equations and
the standard "all or none" penetration criterion to calculate the hourly
51
-------
cr
UJ
00
CD
O
Q
LJ
I
o
e •
D
ON1
O
in
cs
o
CN
_ o
I
e
0)
o u
e o
O <4H
y
y^N
C 01
C3 C
0) O
S -H
J-l
X «
•O
4)
CU .-<
CO tO
o i-i
0)
c o
%*•*
QTJ 1)
CO SO
4J 03 .
<8 SJ -O
<-l CU O
3 > -H
y co )-i
i-l ^ (U
co a
o CN
o e
u-i CO O
O -rt
4-1
3
in
C 0
o
en ca
•* C CO
I-l O >
a) T-I a)
a, u
S a cu
o u £
CM
CN
IV
(-1
3
00
O
(O
o
m
o
-<*•
NOLLVclLN30NOO
52
-------
Q
UJ
>
ce
UJ
00
CD
O
<
CN
O
Q 00
Z)
o
O
b
-------
Q
UJ
>
o:
UJ
oo
GO
o
Q
UJ
o
o
D
CM
O
C/)
o
(N
NOIiVdlN30NOO
54
-------
SC>2 GLC for two days. Except for transition periods with sharp changes in
stability and mixing height, no significant differences are noticeable in
the calculated values. This is probably due to the fact that area sources
are significant contributors to the SC>2 GLC in Philadelphia. Furthermore,
the stack heights of point sources in the inventory are fairly low (under
100 m) and the plumes are not very buoyant. These plume conditions
together with the high mixing depths input to the model did not give
complete penetration. This limited testing of the the new plume rise and
penetration schemes in this evaluation, therefore, did not permit a
conclusive assessment of these methods. We would need a more suitable and
better-defined data set for this purpose.
SULFATE
The scatterplot of calculated versus observed 12-hour average fine
sulfate concentrations is shown in Fig. 25. The linear regression fit is
also shown in the figure. The statistics for this evaluation are given in
Table 4.
The mean of ratios of calculated and observed values, V±/0±, is 1.45;
the correlation coefficient is 0.77 and the slope is 0.84 both day and
night. Since the background is estimated to be a large (25% or more) part
the observed GLC, this high correlation suggests that the background
concentrations, arising from long-range transport and regional inflow of
the species across model boundaries, may be decisive in determining the
particulate sulfate levels in urban areas. The high value of 0.87 for the
index of agreement both day and night suggests that the background
55
-------
S04 CCNCEUfiiATIGH i /iO/a3 i
LESESD: A = i OSS. 3 = 2 OSS, ETC.
SO *•
C 40
A
L
C
ij
I
A 30
T
E
D
iO
A
A A AA
A A A
A AA A A
A AAA
A A
A A
A A A A A Ax-lf A
-A A A A A 3 ft A 3
Aft A 3 AAX A A
A
A A 3BAB XA AA A
AAA 3CE/SCA
A 4A3 A C5S2GBBA AAABfi A A
A A AA A,%{AAB
A A JjXfeABA 3 A A 3 AA
A 3C£2f3E3DC AA A A A3
AB AAA A 3 A
ELC B 3
3 A DA A A
A
10
20
:o
OBSERVED
40
Figure 25. Comparison of calculated and observed 12-hour average
sulfate concentrations for the evaluation period.
56
-------
TABLE 4
Summary of PEM-2 Evaluation Statistics for Sulfate.
Observations
Range
Mean
S.D.
R
Slope
Intercept
Mean of (PI/OI)
S.D. of (Pi/Oi)
Bias
S.D. of Difference
Average Absolute Gross Error
RMSE
Index of Agreement
Mean Fractional Error
MSE(u)
MSE(s)
MSE(u)/MSE
MSE(s)/MSE
Total
Obs. Calc.
307
0-33 0-36
9.85 11.17
6.91 7.54
0.77
0.84
2.87
1.45
1.89
-1.31
4.91
3.42
5.08
0.87
-0.42
22.87
2.91
89%
11%
Day
Obs. Calc.
152
0-29 1-32
9.19 10.18
6.04 6.58
0.77
0.84
2.48
1.46
2.37
-1.00
4.32
2.86
4.42
0.87
-0.31
17.61
1.94
90%
10%
Night
Obs. Calc.
155
0-33 0-36
10.51 12.13
7.63 8.24
0.77
0.84
3.36
1.44
1.25
-1.62
5.43
3.97
5.65
0.87
-0.52
27.67
4.21
87%
13%
Note; The units of Range, Mean, S.D. , Intercept, Bias, S.D. of
Difference, Average Absolute Gross Error, and RMSE are
57
-------
concentrations determined as described earlier may be appropriate.
The sulfate residuals are plotted against the observed concentrations
in Fig. 26. No clear bias in predictions is evident in the results. The
mean difference over the entire range of concentrations is -1.31 Mg/m^,
i.e., the model slightly overpredicts the sulfate GLC. The RMSE is 5.08
Vg/rar, and about 89 percent of the MSE is unsystematic. This suggests
that PEM-2 performs well in simulating the PAFS sulfate concentrations
resulting from the background contributions and primary emissions of the
species, and chemical transformation of SC>2.
In order to investigate the role of background concentrations in the
correlation between the observed and calculated GLC, a stepwise regression
analysis has been performed. This consists of fitting regression relations
between the observed GLC values as dependent variables, and the background
and model-calculated concentrations as independent variables in a stepwise
procedure, as follows:
Step 0 : Y = aL + eL
Step 1 : Y = a2 + b2 X + e2
Step 2 : Y = a3 + b_ X + c_ Z + e_
where Y = 0^ , the observed ground-level concentrations,
X = B£ , the estimated background concentrations,
Z = PI - E± , the model-calculated concentrations,
e. = the residual errors, (k = 1, 2, 3),
and a^ , b, , and c, are the regression coefficients. At each step, the
residual error, e, and correlation coefficient, R, are calculated to see
if a reduction in the sum of squared errors (SSE) and an increase in the
58
-------
3Qi CONCEHTRATIGK
LE3END: A = 1 OSS. 3 = 2 OSS, ETC.
30 *
20
10
E
S
I
D 0
U
A
L
-10
-20
A A
A A A
A A A A A A A
A A
A A 3 4 A A
3A A A A AA
A A A A 3 ; A A A A
AAS AAriAB 3 A A A A AA A 3A A 4 A A
M-AA"AIBC-fLC5CA3CAA--A-AC3ABAAS--3—A-A-AA A—2 A AA
FCBDEBCE3 A AB2D83A8SACS E A A AA A A A
A 38AEAA A3 A AA A 3BABAA A A A A A A A
A A AA A A AAA3 A A A AA A A A A A AA
A -AA A AA A AA A
A A A
A A A
A A A A
A A
A A
A A
-30 +
0 10 20 30 40
OBSERVES C2HCENTRATIOM
Figure 26. Sulfate residuals (Di = 0^ - P^) versus observed sulfate
concentrations for the evaluation period.
59
-------
square of correlation coefficient (RSQ) occurs with the increase in the
number of independent variables in the regression.
The stepwise regression equations, and the corresponding values of RSQ
and SSE, are determined separately for the total, daytime, and nighttime
observed sulfate GLC data, as shown in Table 5. The coefficients for the
model-calculated concentrations in the regression equations are smaller
than those for the background concentrations. This indicates that the
latter account for a larger fraction of the correlation between the
observed and predicted GLC than the local source contributions calculated
by the model. This is particularly noticeable at nighttime, thus
suggesting that either a major portion of the observed concentrations is
due to the background, or that the model is not geared towards estimating
the GLC at night. The decrease in SSE from step 0 to step 1 is much larger
than that from step 1 to step 2 in all cases, especially at night. This
confirms that background concentrations play a more important role at night
for sulfate.
Figure 27 shows a comparison of the calculated and observed daily mean
concentrations of sulfate (averaged over all six PAFS stations) for each of
the 29 evaluation days. There is excellent agreement between the calcu-
lated and observed concentrations. If the results are plotted separately
for 12-hour averages, the differences in model performance between daytime
and nighttime (Figs. 28 and 29) are seen to be small.
60
-------
TABLE 6
Results of Stepwise Regression Analysis for Sulfate.
RSQ SSE
Total Data (sample size = 307)
Step 0 : Y = 9.85 , - 14608
Step 1 : Y = 2.98 + 0.83 X , 0.59 6056
Step 2 : Y = 2.08 + 0.80 X + 0.40 Z , 0.62 5585
Daytime Data (sample size = 152)
Step 0 : Y = 9.19 , - 5506
Step 1 : Y = 3.15 + 0.75 X , 0.51 2703
Step 2 : Y = 2.00 + 0.73 X + 0.60 Z , 0.59 2236
Nighttime Data (sample size = 155)
Step 0 : Y =* 10.51 , - 8968
Step 1 : Y = 3.02 + 0.88 X , 0.64 3249
Step 2 : Y = 2.52 + 0.86 X + 0.19 Z , 0.64 3199
61
-------
Q
UJ
UJ
CO
CD
O
LU
O
D
NOIiV&LN33NOO
62
-------
O
o:
UJ
c/)
m
O
UJ
I—
I
o
O
UJ
(f)
UJ
z:
in
V
sc
O
f
O
-------
ce
UJ
(/>
oa
O
UJ
5
D
O
<
O
a
LJ
s
ID
00
m
cs
01
M
cd
(B
O
es
O
cs
60
i-4
C
U
O
o.
sg 3 «
Q (U C
o
«M a)
SO C
•H «
fc. O
a
c o
iJ O
3)
-------
FINE TOTAL MASS
Figure 30 shows a comparison of all of the calculated and observed
12-hour average fine total mass concentrations for the model evaluation
days. The solid line shows the linear regression fit with a slope of 0.79
and intercept of 10.91 Ug/ra .
The evaluation statistics of paired comparisons for fine total mass are
summarized in Table 6. The high values of the correlation coefficient
(0.75) and the index of agreement (0.84) indicate that the model performs
well in predicting observed concentrations. However, the invariance of
these values both day and night suggest that the background concentrations
may be more important than local source contributions in estimating the FP
concentrations in Philadelphia.
The fine total mass residuals are plotted against the observed con-
centrations in Fig. 31. A tendency towards overprediction is evident in
the plot. The bias over the entire range of comparison is -4.7 yg/m3. The
nighttime bias (-5.39 Pg/m3) is higher than that during the daytime (-4.01
Mg/m3). The RMSE is 9.98 ug/m3, and the major part (71 percent) of the MSE
is unsystematic error. This suggests that PEM-2 performed satisfactorily
in simulating the 12-hour average GLC of FP total mass in the PAFS data.
In order to investigate the importance of FP background concentrations in
Philadelphia, a stepwise regression analysis (similar to that described for
sulfate) has been performed. The regression equations, RSQ, and SSE for
each step are shown in Table 7. These results show that the FP background
concentrations play a major role in correlating to the observed GLC, and
the model-calculated concentrations are only of secondary importance.
65
-------
70
50
A
C
U 40
A
T
£
a 30
20
10
,-;NE TOTAL MSS C^CCNTRATIGSS i
LEEEHD: A = i 23S. 3 = 2 SSS. ETC.
A A A
A A
AA A AA
A A
A A A AA AA
A A AA AA A A / 8 A
A A AA AA A
A A A 3 /ft A
AB A //A A
A A A A A A A M
A A A A/ A
ft AA /3 A
A AAA A
A3 4 A A A JT A A A A
A A A A A 8 3AAXA A AAA
A A AACSAyf ABA3 AAA A
A AAAA/K3AABBC AA
Aft A A A C AA^S A C A A A
AAAAA B 3B^S 9A A
A AAAABB A CA/4AA 3 A A
3A AA/S BAB S A AA A A A
3 X ABABA8A 3 C AA
'AC A A AAA A A A
AAAA A A A A
'A AAA
A A A
A A A
AAA
10
Figure 30.
OBSERVED
Comparison of calculated and observed 12-hour average
FP total mass concentrations for the evaluation period.
-------
TABLE 6
Summary of PEM-2 Evaluation Statistics for Fine Total Mass.
Observations
Range
Mean
S.D.
R
Slope
Intercept
Mean of (PI/OI)
S.D. of (Pi/Oi)
Bias
S.D. of Difference
Average Absolute Gross Error
RMSE
Index of Agreement
Mean Fractional Error
MSE(u)
MSE(s)
MSE(u)/MSE
MSE(s)/MSE
Note: The unit of Range, Mean,
Total
Obs. Gale.
301
7-65 11-76
29.84 34.54 2
12.15 12.81 1
0.75
0.79
10.91
1.24
0.41
-4.70
8.82
7.43
9.98
0,84
-0.85
71.14
28.49
71%
29%
S.D. , Intercept
Day
Obs. Gale.
150
9-59 11-66
9.34 33.35
1.34 11.83
0.75
0.78
10.46
1.20
0.37
-4.01
8.24
6.49
9.14
0.84
-0.74
61.28
22.27
73%
27%
, Bias, S.D.
Night
Obs. Gale.
151
7-65 11-76
30.33 35.72
12.93 13.65
0.75
0.80
11.56
1.27
0.45
-5.39
9.33
8.37
10.75
0.83
-0.97
79.67
35.93
69%
31%
of
Difference, Average Absolute Gross Error and RMSE are
67
-------
FINE TOTAL 1ASS CGHC3TRATIQSS i pc/s3 )
LEacfiD: 4 = ! 332, 3 = 2 033, £TC.
50 1-
+0
30
s
E
5
I 10
s
u
A
L 0
-20
-30
-40
A A
A A A
A A A A A
A A AA A A A 3 3 A A A
A 4 3 3A 3 AAB AA 4 A A A A 3
£ a-Afl--ACAA-3-A-3-4-AA-3-3SftB-3ft3-A—A A-AA—A A-A
A 3AAAA AB 3CA 3ft 3ft EBCAA M A 4 A A A 3 A A A
A 3B Aft 8 BC AA A AA DAAAAA AA 3A A A A3 3 AA
A A 3A 3AA 3B C 3A 3 AA A A Aft A A
A AA 8 3 3 AftA A 3 A AAB A A AA AA AA AA 4
A AftA AftAAA A A 4 A ftA A
AAA AAA Aft Aft A AAA
A ft A 8 ft A A
A A AA A A
A A 3
A A
A
10 20 30 40
OBSERVED CGNCEHTRATIOH
Figure 31. FP total mass residuals (D^ = 0^ - P^) versus observed
12-hour average FP concentrations for the evaluation
period.
68
-------
TABLE 7
Results of Stepwise Regression Analysis for Fine Total Mass,
RSQ
SSE
Total Data (sample size = 301)
Step 0 : Y = 29.84 ,
Step 1 : Y = 8.55 + 0.77 X ,
Step 2 : Y = 6.03 + 0.78 X + 0.36 Z ,
Daytime Data (sample size = 150)
Step 0 : Y = 29.34 ,
Step 1 : Y - 9.00 + 0.73 X ,
Step 2 : Y = 6.85 + 0.73 X + 0.39 Z ,
Nighttime Data (sample size = 151)
Step 0 : Y = 30.33 ,
Step 1 : Y = 8.30 + 0.81 X ,
Step 2 : Y = 5.69 + 0.81 X + 0.31 Z ,
0.60
0.62
0.66
0.67
44316
17634
16895
—
0.54
0.55
19158
8771
8539
25084
8602
8250
69
-------
Figure 32 shows that the model tracks the observed daily mean
concentrations (averaged over the six stations) fairly well for the
evaluation period. Similar comparisons for the daytime and nighttime
12-hour average concentrations (Figs. 33 and 34, respectively) show no
significant differences in performance.
COARSE TOTAL MASS
Figure 35 shows a comparison of the calculated and observed 12-hour
average coarse total mass concentrations for the 29 evaluation days. The
solid line shows the linear regression fit. The slope is 0.14 and the
intercept is 13.45 Ug/m3.
The model evaluation statistics for coarse total mass are summarized in
Table 8. The mean of ratios, Pi/Oi, is 1.29 and the corresponding standard
deviation is 0.81. The correlation coefficient of 0.25 indicates a large
degree of randomness in the paired comparison of coarse mass concentra-
tions. The relatively low index of agreement (0.47) suggests that the
predictions of CP contain more error than the results for FP and sulfate.
The coarse mass residuals are plotted against the observed concentra-
tions in Fig. 36. The model shows a clear bias to overpredict the observed
concentrations below 20 yg/m3, and underpredict 0^ above this value. The
mean of differences over the entire range of concentrations is -0.83 pg/m3,
i.e., the model is slightly conservative. The average absolute gross error
of 5.74 Pg/m3 is much less than the mean of observed concentrations. The
RMSE is 8.28 Ug/m3, but 73 percent of the MSE is systematic. This differs
from the results for sulfate and FP for which a major part of the MSE is
unsystematic. This seems to suggest that there is room for improvement in
70
-------
Q
Ld
en
QD
o
UJ
§
Z5
O
<
o
* •
a
CO
CO
o
UJ
NOLLVdiN30NOO
71
-------
Q
UJ
>
a:
UJ
oo
m
O
Q
UJ
^
O
O
d
OO
•3-
cs
O
0-4
•<
>30
cd
O
1)
CD
13
O .
oj en
JJ O
O. T-l
4) U
a «
X i-
0) i-i
e
. aj
-------
Q
cr
LU
(/)
DQ
O
00
GO
o o
UJ H-
< LJ
—1 ~Z.
o C
CJ
* •
o
O
CM
M
(0
O)
>
CO
O
0)
eo
•*
(3
O •
IM 0}
c
UJ O
O. -H
01 JJ
CJ CO
X M
01 4J
C
.
-------
so
C 40
A
L
C
U
L
ft 30
T
E
3
20
to
0 «•
COASSE rOTAL S«SS CONCESTRfiriSffS ! pg/*3 i
LESENO: A- = 1 QSS, 3 = 2 3BS. ETC.
3
A A AA
A A A
A AA AA A AAAA i
A A 3 A A A
AAAA A 43 A AB A A
A A3 3 AA 3AAAAA 3 A A A
A A CCAA8FSSCSCAA 3 A
A A 3 A
TABC c CECFcsec 3 CCSAA A
AA8AAABBCCF8C A AAAA A
AEAAC3 H AA A AA A A A A
A ASA A A3 A
A
A A
10
50
40
OBSERVES
Figure 35. Comparison of calculated and observed 12-hour average
CP total mass concentrations for the evaluation period.
74
-------
TABLE 8
Summary of PEM-2 Evaluation Statistics for Coarse Total Mass.
Observations
Range
Mean
S.D.
R
Slope
Intercept
Mean of (PI/OI)
S.D. of (Pi/Oj.)
Bias
S.D. of Difference
Average Absolute Gross Error
RMSE
Index of Agreement
Mean Fractional Error
MSE(u)
MSE(s)
MSE(u)/MSE
MSE(s)/MSE
Note: The unit of Range, Mean,
Total
Obs. Calc.
298
1-54 5-31
14.62 15.44 1
8.15 4.46
0.25
0.14
13.45
1.29
0.81
-0.83
8.25
5.74
8.28
0.47
-0.36
18.57
49.99
27%
73%
S.D., Intercept
Day
Obs. Calc.
148
1-47 7-26
3.39 14.89
6.58 3.63
0.25
0.14
13.05
1.35
0.98
-1.50
6.68
4.90
6.82
0.47
-0.52
12.27
34.26
26%
74%
, Bias, S.D.
Night
Obs. Calc.
150
3-54 5-31
15.84 16.00
9.31 5.10
0.23
0.13
14.01
1.22
0.59
-0.16
9.53
6.56
9.50
0.45
-0.20
24.48
65.82
27%
74%
of
Difference, Average Absolute Gross Error, and RMSE are Pg/m3,
75
-------
50 *
s :o
10
-20
COARSE TOTAL JMSS CONCENTRATISNS i |ig/«3
LtSEHO: A s 1 035. 3 * 2
-------
modeling the CP concentrations, for example, by using a different set of
values for the deposition and gravitational settling velocities for
particles of size 2.5 to 10 microns.
In order to investigate the role of background concentrations in the
comparison between the observed and calculated GLC, a stepwise regression
analysis has been performed for the total, daytime, and nighttime CP data
separately. The regression equations, RSQ, and SSE for each step are shown
in Table 9. The results show that the correlation coefficients for both
background and model-calculated concentrations are of the same order,
indicating that each accounts for about the same fraction of the
correlation between the observed and predicted GLC. The decrease in SSE is
more significant from step 0 to step 1 than from step 1 to step 2.
However, this decrease is not so significant as that for the sulfate or FP
species.
Figures 37, 38, and 39 show the model performance in predicting daily
mean, daytime and nighttime (12-hour average) coarse total mass concentra-
tions, respectively. All concentrations are averaged over the six PAFS
stations. The overall agreement between the calculated and observed
variations is good with no discernable difference between the daytime and
nighttime performance. Though the local source contributions may be
relatively more important for the coarse mass than for sulfate and fine
mass, the background plays an important role in determining the overall CP
concentration levels in Philadelphia.
77
-------
TABLE 9
Results of Stepwise Regression Analysis for Coarse Total Mass.
RSQ
SSE
Total Data (sample size = 298)
Step 0 : Y = 14.62 ,
Step 1 : Y = 6.38 + 0.70 X, 0.12
Step 2 : Y = 3.84 + 0.72 X + 0.67 Z, 0.16
Daytime Data (sample size = 148)
Step 0 : Y = 13.39,
Step 1 : Y = 6.25 + 0.61 X, 0.11
Step 2 : Y - 2.98 + 0.70x + 0.71 Z, 0.15
Nighttime Data (sample size = 150)
Step 0 : Y = 15.84,
Step 1 : Y = 6.99 + 0.75 X, 0.13
Step 2 : Y = 5.23 + 0.73 X + 0.50 Z, 0.15
19723
17264
16595
6364
5659
5383
12908
11166
10950
78
-------
Q
UJ
>
a;
UJ
c/>
CD
O
t/)
O
i—
UJ
C/1
CE
<
O
O
O
a
o
ro
CM
o
CM
l-< l-C
to o
CO
a, c
o o
(8 to
HI 4J
S ca
cu
- *
Q 13
4J
to
(—1
3
u
1— 1
to
U-l
O
C
o
en
•^
i_j
to
c.
6
O
0
to
J-l
CD
>
O
•a
0)
50
CO
01
to
tn
c
o
•^
to
J-l
u
c
0)
y
c
0
a
ca
tn
to
g
«
*o
o
^J
(JJ
a.
e
o
*^i
^j
CO
3
r-l
to
>
0)
a)
^
u
3
bO
O
CM
NOIlVdIN30NOO
79
-------
UJ
>
OS
UJ
C/)
00
o
_J
z>
o
<
o
a
C/)
(/)
<
2
£
o
UJ
en
o
o
o
to
cs
o
cs
60
eo
14
3
O
0)
E
03
T3
O .
UJ 9)
C
•u O
y «
x <->
01 4J
c
„ a)
r>- O
m e
o
• o
bC
•H 01
Bb 0}
-------
Q
UJ
UJ
en
CD
O
(S)
in
00
°z
<
o
n
NOIiVaiN30NOO
81
-------
SECTION 5
CONCLUSIONS
This report described an evaluation of the Pollution Episodic Model
Version 2 using 29 days of the Philadelphia Aerosol Field Study data. This
evaluation is designed to test the model performance by comparing its
concentration estimates for four pollutant species (S02, sulfate, fine and
coarse total mass) to the measured air quality data, using appropriate
statistical measures of performance.
The emphasis in this evaluation is on primary fine and coarse inhalable
urban particulate matter, as well as pollutants that undergo chemical
transformation in the atmosphere to form fine secondary particulates (e.g.,
S02 to sulfate aerosol). For the evaluation period, PEM-2 predicted
12-hour average concentrations of S02, sulfate, fine and coarse total mass
to within a factor of two, which is the best that may be expected
considering the natural variability in model input data (Hanna, 1981).
The hourly S02 concentration data are considered to provide a demanding
test of the performance of PEM-2, especially since the background S02
concentrations are assumed to be zero. The model performance for S02 is
better during daytime than at nighttime, and generally better at the
suburban stations, which are affected mostly by point sources, than at
downtown stations. The latter are impacted heavily by the urban area
sources, which are thought to be major contributors to the errors in the
82
-------
concentration estimates. These errors result from the difficulty in
characterizing area source emissions, lack of information on source
conditions such as release height, plume rise, and building-wake effects,
as well as the narrow plume concept and other simplifying assumptions used
in the concentration algorithms for area sources.
The calculated and observed SO concentrations are closer at all PAFS
stations during daytime than at night. The model generally overestimates
the hourly concentrations at night, and underpredicts during the day at
rural locations which are not significantly impacted by area sources. This
stability bias of the model is attributed to the the problems associated
with modeling low-level urban sources, and the model sensitivity to the
characterization and partition of emissions between point and area sources.
The model performance for the particulate species is good; both the
12-hour and daily mean concentrations (averaged over the six PAFS stations)
over the evaluation period are predicted well. For sulfate and fine total
mass, the values of the correlation coefficient and index of agreement are
high, and a large part of the estimated MSE is unsystematic. For coarse
total mass, the evaluation statistics are not as good by comparison, and
the large systematic error suggests there is room for model improvement.
On the average, the model estimates of concentrations for the particulate
species are slightly conservative (overpredictions).
The overall good performance of the model for particulate species
should be interpreted with caution since the background concentrations,
resulting from long-range transport and regional inflow of species across
model boundaries, far exceed the urban source contributions to the surface
83
-------
concentrations in Philadelphia. When the wind direction changes
significantly, no particular source contributes longer than one hour to
receptor concentrations in PEM-2 (with the possible exception of receptors
located within area sources). On the other hand, the determination of
background concentrations from PAFS data is somewhat subjective due to the
small number of monitoring stations available. Ideally, the background
monitors should be located on the calculation grid boundaries, away from
the influence of sources being modeled. Nevertheless, judging from the
model performance in this evaluation, the values of background particulate
concentrations estimated from observed GLC at suburban monitors and wind
direction analysis for each 12-hour averaging period seem appropriate.
It appears that the background particulate concentrations are important
not only in Philadelphia, but in other urban areas as well. Wolff et al.
(1985) studied the influence of local and regional sources on the
concentrations of inhalable particulate matter at four (urban,
industrialized, suburban, and rural) sites in southeastern Michigan, and
found that FP was dominated by regional influences rather than local
influences at all four sites, while CP was dominated by local sources. The
regional influences were most pronounced on the sulfate levels which
accounted for the largest fraction (40 - 50%) of the FP in their study.
It may be possible to establish the relative importance of the local
and regional contributions to the urban FP and CP concentrations from an
analysis of the National Inhalable Particulate Network data. In a
preliminary interpretation of these data, Pace and Rodes (1981) concluded
that CP and FP concentration levels are substantially different among urban
areas and, on an urban scale, FP averages are generally more homogeneous
84
-------
indicating a regional pattern; the lower degree of uniformity in CP
(compared to FP) concentrations in urban areas may be due to a larger
contribution of CP from local sources. The results of the present study
seem to support this conclusion.
The characteristics of emissions in Philadelphia appear to be
substantially different from those in St. Louis. This is demonstrated in
Table 10 which compares the daily average emission rates of the species
from point and area sources during the PAFS and the RAPS experiments.
Unlike in St. Louis, the area sources in Philadelphia are significant
contributors of S02 and sulfate, and point source contributions to fine and
coarse total mass are very important. Because of the differences in
emission characteristics between different urban areas, it is desirable to
evaluate PEM-2 with as many detailed data sets as possible. In an
evaluation of PEM using St.Louis/RAPS data, Pendergrass and Rao (1984)
attributed the large overpredictions of FP and CP by the model to
overestimation of emission rates of area sources in the inventory and their
incorrect location, among other factors.
The meteorological data used in this evaluation are surface observa-
tions from NWS station at the Philadelphia Airport (PHL). These data are
only approximations to real conditions in the Philadelphia urban area.
Errors in wind direction may cause the model to impact particular receptors
which may be completely ignored in reality. Errors in wind speed and
stability classification may significantly affect the diffusion, plume rise
and penetration calculations. It is well known that the frequency and
intensity of nocturnal inversions over cities are decreased due to enhanced
thermal and mechanical mixing resulting from urban heat island and
85
-------
TABLE 10
Comparison of Dally Average Emission Rates (kg/s) from Area and
Point Sources in Philadelphia and St. Louis.
Pollutant
Area Sources
Point Sources
S02
Sulfate
Fine Total Mass
Coarse Total Mass
Philadelphia/PAFS
15.674
0.603
5.459
3.400
51.561
1.639
3.997
1.637
S02
Sulfate
Fine Total Mass
Coarse Total Mass
St. Louis/RAPS
0.434
0.008
2.518
6.631
36.018
0.266
0.211
0.921
86
-------
increased roughness effects. Hence, the diffusion conditions in the lower
atmosphere over cities tend to be of near-neutral type, without the strong
diurnal variations (see Fig. 4) found at airport locations. The plume
dispersion parameters used in PEM-2 account for the enhanced turbulence and
mixing over urban areas by using greater sigraa values (than in the open
country) for given stability class. However, due to lack or unreliability
of on-site meteorological data, it cannot be ascertained if the actual
dispersion conditions over Philadelphia are adequately simulated in this
study.
The effective height of area-source emissions, taken as 10 m (typical
of building-top releases) in this evaluation, is an approximate value used
to mitigate the large GLC calculated at receptors located within the area
sources. Better characterization of area and point source emissions would
be helpful. The influence of the new plume rise equations and new
plume-penetration schemes used in this evaluation (for buoyancy-dominated
point-source plumes in unstable/neutral atmosphere) could not be assessed
since no significant differences are noticeable in the hourly S02 GLC
calculated for two days from the new schemes and the standard schemes,
except for transition periods with sharp changes in stability and mixing
height. However, these results are not conclusive, and more tests with
suitable data are necessary to bring out the differences in performance
between the various schemes.
The primary objective of this study is to evaluate the performance of
PEM-2 with emphasis on estimating concentrations of urban particulate
matter. Based on the results, we conclude that PEM-2 is able to simulate
the observed concentrations of S02, sulfate, FP and CP total mass in PAFS
87
-------
data fairly well. The results of stepwlse regression analysis in this
study suggest that the observed GLC of the urban particulate species are
highly correlated to their estimated background concentrations. The latter
are significantly larger than the local source contributions (as calculated
by the model) in Philadelphia and, therefore, play an important role in
predicting the GLC of particulate matter. With the PAFS data, we could not
readily identify where further improvements can be made to the model; this
would require a more extensive and better-defined evaluation data set.
The background concentrations of particulate matter in an urban area
are determined by the primary emissions, chemical transformation, and
transport on scales much larger than the city size. The background
concentrations in Philadelphia are large because of its proximity to other
densely-populated urban and industrial centers in the eastern U.S. The
relatively smaller contributions from the local sources of particulate
matter are determined essentially by the given emissions and meteorology.
Therefore, additional effort should be directed towards an examination of
the emissions characterization, input meteorology, and variability of
background concentrations. An objective methodology to determine the
background particulate concentrations should be evolved and the monitoring
stations in the field programs in future should be located accordingly.
88
-------
REFERENCES
Engineering—Science, 1984: Development of an emission inventory for
urban particle model validation in the Philadelphia AQCR. Contract
No. 68-02-3509, U.S. Environmental Protection Agency, Research Triangle
Park, NC, 613 pp.
Fox, D. G., 1981: Judging air quality model performance. Bull, of Amer.
Meteor. Soc. 62, 599-609.
Fox, D. G., 1984: Accuracy in air quality modeling. Bull, of Amer. Meteor.
Soc. j>5_, 27-36.
Hanna, S. R. , 1981: Natural variability of observed hourly SC>2 and CO
concentrations in St. Louis. Atmos. Environ. 16, 1425-1440.
Pace, T. G., and C. E. Rodes, 1981: Preliminary interpretation of
Inhalable Particulate Network data. Paper No. 81-5.2, Vol. I,
Proc. of 74th Annual Meeting of APCA, Philadelphia.
PEDCO Environmental, Inc., 1983: The 1982 Philadelphia Aerosol Field
Study. Data Collection Report. Contract No. 68-02-3509, U.S.
Environmental Protection Agency, Research Triangle Park, NC.
Pendergrass, W. R., and K. S. Rao, 1984: Evaluation of Pollution
Episodic Model using RAPS data. EPA-600/3-84-087, U.S.
Environmental Protection Agency, Research Triangle Park, NC;
NOAA Tech. Memo. ERL ARL-128. 47 pp. [NTIS PB84-232537],
Ray, A. A., 1982: SAS User's Guide: Statistics, 1982 Ed.
SAS Institute, Inc., Gary, NC, 584 pp.
Rao, K. S., 1983: Plume concentration algorithms with deposition,
sedimentation, and chemical transformation. EPA-600/3-84-042, U.S.
Environmental Protection Agency, Research Triangle Park, NC; NOAA
Tech. Memo. ERL ARL-124, 87 pp. [NTIS PB84-138742].
Rao, K. S., and M. M. Stevens, 1983: Pollution Episodic Model User's
Guide. EPA-600/8-84-008, U.S. Environmental Protection Agency,
Research Triangle Park, NC; NOAA. Tech. Memo. ERL ARL-125, 186 pp.
[NTIS PB84-164128].
Rao, K. S., 1985: User's Guide for PEM-2: Pollution Episodic Model
(Version 2). EPA-600/8-85- , U.S. Environmental Protection Agency,
Research Triangle Park, NC; NOAA Tech. Memo. ERL ARL-137, 128 pp.
Available from NTIS.
Rao, S. T., G. Sistala, V. Pagnotti, W. B. Petersen, J. S. Irwin, and
D. B. Turner, 1985: Evaluation of the performance of RAM with the
Regional Air Pollution Study data base. Atmos. Environ. 19,
229-245.
89
-------
Ruff, R. E., 1983: Application of Statistical methods to diagnose
causes of poor air quality model performance. Atmos. Environ. 17,
291-297.
Texas Air Control Board, 1979: User's Guide: Texas Episodic Model.
Permits Section, Austin, Tx, 215 pp.
Turner, D. B., and J. H. Novak, 1978: User's Guide for RAM.
EPA-600/8-78-016a, U.S. Environmental Protection Agency, Research
Triangle Park, NC, Vol. 1, 60 pp.
Turner, D. B., and J. S. Irwin, 1985: The relation of urban model
performance to stability. Air Pollution Modeling and Its
Application IV, C. De Wispelaere, Ed., Plenum Press, New York,
721-732.
Wolff, G. T., P. E. Korsog, D. P. Stroup, M. S. Ruthkosky, and M. L.
Morissey, 1985: The influence of local and regional sources on
the concentration of inhalable particulate matter in southeastern
Michigan. Atmos. Environ. 19, 305-313
90
------- |