United States
Environmental Protection
Agency
Office of Air Quality ... .„_.,.
Planning and Standards
Research Triangle Park, NC 27711
EPA-454/R-92-004
August 1992
Air
& EPA
EVALUATION OF CO
INTERSECTION MODELING
TECHNIQUES USING A
NEW YORK CITY DATABASE
-------
-------
EPA-454/R-92-004
Evaluation of CO Intersection Modeling
Techniques Using a New York City Database
U.S. Environmental Protection Agency
Office of Air Quality Planning and Standards
Technical Support Division
Research Triangle Park, NC 27711
August 1992
-------
Notice
This report has been funded by the United States Environmental Protection Agency
under contract 68D90067 to Sigma Research Corporation. Thomas N. Braverman served as
the EPA work assignment manager. Any mention of trade names or commercial products is
not intended to constitute endorsement or recommendation for use.
11
-------
Table of Contents
List of Tables v
List of Figures
Acknowledgements xv
1.0 Introduction i
1.1 Overview of the Analysis i
1.2 Study Objectives 2
1.3 .Report Organization 3
2.0 Intersection Modeling Techniques 5
2.1 Overview of Modeling Techniques 5
2.2 Model Summaries 6
2.2.1 EPA Intersection (EPAINT) Model 11
2.2.2 FHWA Intersection (FHWAINT) Model 12
2.2.3 VOLUME9/MOBILE4 (VOL9MOB4) '. 12
2.2.4 Georgia Intersection Model (G3M) 13
2.2.5 CAL3QHC 14
2.2.6 CALINE4 15
2.2.7 TEXIN2 16
2.2.8 Intersection Midblock Model (TMM) 16
3.0 The New York City Database 19
3.1 Description of the Six New York Intersections 19
3.2 Description of the Data Collected 27
3.3 Analysis of the Observations 31
3.3.1 Analysis of the Wind Structure 31
3.3.2 Characterization of the Wind Speed and
Stability Class for Modeled Hours 40
3.3.3 Traffic Counts 40
4.0 Modeling Methodology 45
4.1 Model Input Data 45
4.1.1 Intersection Configurations 45
4.1.2 Traffic and Emissions Characterization 54
4.1.3 Meteorological and Background Data 67
4.2 Dispersion Modeling Techniques 68
in
-------
Table of Contents (Continued)
Page
5.0 Statistical Evaluation Protocol 71
5.1 The Model Evaluation Support System (MESS) 71
5.1.1 Paired Statistics 71
5.1.2 Unpaired Statistics 73
5.1.3 MESS Analysis Products 73
5.2 Model Evaluation Scoring Scheme 75
5.2.1 Screening Test : 75
5.2.2 Refined Evaluation 76
5.2.3 Summary of Scoring Scheme 83
5.2.4 Limitations of the Scoring Scheme 84
6.0 Model Performance Results 85
6.1 Phase I Results: 8 Models/6 Sites Using MOBILE4.0
Emissions . 85
6.1.1 Paired and Unpaired Statistics 85
6.1.2 Screening Results 92
6.2 Phase E Results: 5 Models/3 Sites Using MOBILE4.1
Emissions 106
6.2.1 Paired and Unpaired Statistics 106
6.2.2 Diagnostic Analysis 123
6.2.3 Regulatory Worst-Case Analysis 127
6.2.4 Scoring Scheme Results 130
7.0 Summary and Conclusions 151
8.0 References 155
Appendix A Additional Phase I MOBILE4.0 ANALYSES
Appendix B Residual Plots Using MOBILE4.1 Emissions
-------
List of Tables
Table
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Intersection Model Input Requirements
Summary of the Available Route 9A Reconstruction Project Monitoring
Data „
CO Probe Heights (feet) for Each Monitor „
Summary of Available Hourly Traffic Data for Each Intersection
Segment
Tabulation of Stability Classification by Site
Tabulation of Wind Speed by Site ..'..."
Summary of Intersection Configurations
Summary of Saturated Flow and Percent Red Time
Traffic Cruise Speeds (mph) Used in Modeling Analysis
Mileage Accumulation Rates and Registration Distributions
Used in Modeling Analysis for Ages 1-25
Thermal States (% Cold) Used hi Modeling Analysis
Array of Calculated Performance Measures and Statistics
Paired hi Time and/or Location
Array of Calculated Performance Measures and Statistics
For the "N" Highest (Unpaired) Data Sets (Where N is 25)
Tabulation of Wind Speed/Stability Classification
by Site
Screening Test Results for Site #1 Using MOBILE4.0
Emissions Methodology
Screening Test Results for Site #2 Using MOBILE4.0
Emissions Methodology
... 7
...28
. .. 29
. .. 30
, , , 41
41
, , , 46
...55
, , 58
. .. 62
65
...12
...74
...78
...99
. . . 100
-------
List of Tables (Continued)
17 Screening Test Results for Site #3 Using MOBDLE4.0
Emissions Methodology ............................. . . o
18 Screening Test Results for Site #4 Using MOBELE4.0
Emissions Methodology ......................................... 102
19 Screening Test Results for Site #5 Using MOBILE4.0
Emissions Methodology ......................................... 103
20 Screening Test Results for Site #6 Using MOBILE4.0
Emissions Methodology ............................. 'm ........... jQ4
21 . Summary of EPA Screening Test Results for Each Model Evaluated in
the New York City CO Intersection Modeling Analysis (Using
MOBILE4.0 Emissions Methodology) ............................... 105
22 All Observed and Predicted CO Concentrations (ppm) Paired hi Time
and Location Using MOBILE4.1 Emissions ........................... 107
23 Highest Observed and Predicted CO Concentrations (ppm) Event by
Event (Paired in Time) Using MOBILE4.1 Emissions .................... 109
24 Highest Observed and Predicted CO Concentrations Paired by Station Using
MOBILE4.1 Emissions .......................................... HO
25 25 Highest Predicted and Observed CO Concentrations (ppm) Using
MOBBLE4.1 Emissions .......................................... H3
26 25 Highest Observed and Predicted CO Concentrations (ppm) Unpaired hi
Time or Location Using MOBILE4.1 Emissions ........................ 115
27 Comparison of Top-Ten Observed Concentrations with Predicted
Concentrations Using Regulatory Default and Observed Meteorology
Using MOBELE4.1 .............................................. 12g
28 Robust Highest Concentrations and Fractional Bias by Operational/
Diagnostic Component .......................................... 131
VI
-------
Table
29
List of Tables (Continued)
Summary of the Composite Model Comparison Measures (CM) of Differences
Between Model Performance as Measured by the Absolute Fractional Bias in
Predicting Robust Highest Concentrations for MCM Statistics 150
A-l All Observed and Predicted CO Concentrations (ppm) Paired in Time
and Location Using MOBILE4.0 A-l
A-2 Highest Observed and Predicted CO Concentrations (ppm) Event by
Event (Paired in Time) Using MOBILE4.0 A-4
A-3 Highest Observed and Predicted CO Concentrations (ppm) Paired by
Station Using MOBBLE4.0 A-6
A-4 25 Highest Observed and Predicted CO Concentrations (ppm) Unpaired
in Time or Location Using MOBILE4.0 A-9
A-5 A Comparison of Top-Ten Observed Concentrations with Predicted
Concentrations Using Regulatory Default and Observed Meteorology
Using MOBILE4.0 A-ll
Vll
-------
List of Figures
1 Locations of the six intersections and two background stations in the New
York City database • 20
2 Site #1, West/Chambers, location map. The Battery Park background site is also
shown 21
3 Site #2, 34th/8th, location map. The Post Office background site is also shown . . 22
4 Site #3, 65th/Broadway, location map 23
5 Site #4, 57th/7th, location map 24
6 Site #5, 34th/12th, location map 25
7 Sites #6A, Battery Tunnel, location map 26
8 Cos(9), where 9 is the wind direction difference between Monitors 1 and 2, as a
function of the wind direction at Monitor 1 for all 1-hour average data
at Site #1 33
9 The ratio of the wind speed difference between Monitors 1 and 2 to the average wind
speed at the same monitors as a function of the wind direction at Monitor 1 for all
1-hour average data at Site #1 33
10 Cos(6), where 9 is the wind direction difference between Monitors 1 and 2, as a
function of the wind direction at Monitor 1 for all 1-hour average data
at Site #2 34
11 The ratio of the wind speed difference between Monitors 1 and 2 to the average wind
speed at the same monitors as a function of the wind direction at Monitor 1 for all
1-hour average data at Site #2 34
12 Cos(9), where 9 is the wind direction difference between Monitors 2 and 3, as a
function of the wind direction at Monitor 3 for all 1-hour average data
at Site #2 35
13 The ratio of the wind speed difference between Monitors 2 and 3 to the average wind
speed at the same monitors as a function of the wind direction at Monitor 1 for all
1-hour average data at Site #2 35
viu
-------
List of Figures (Continued)
Figure
14 Cos(6), where 6 is the wind direction difference between Monitors 1 and 2, as a
function of the wind direction at Monitor 1 for all 1-hour average data
at Site #3 35
15 The ratio of the wind speed difference between Monitors 1 and 2 to the average wind
speed at the same monitors as a function of the wind direction at Monitor 1 for all
1-hour average data at Site #3 36
16 Cos(6), where 9 is the wind direction difference between Monitors 1 and 2, as a
function of the wind direction at Monitor 1 for all 1-hour average data
at Site #4 37
17 . The ratio of the wind speed difference between Monitors 1 and 2 to the average wind
speed at the same monitors as a function of the wind direction at Monitor 1 for all
1-hour average data at Site #4 37
18 Cos(9), where 6 is the wind direction difference between Monitors 1 and 2, as a
function of the wind direction at Monitor 1 for all 1-hour average data
at Site #5 38
19 The ratio of the wind speed difference between Monitors 1 and 2' to the average wind
speed at the same monitors as a function of the wind direction at Monitor 1 for all
1-hour average data at Site #5 38
20 Cos(6), where 9 is the wind direction difference between Monitors 1 and 2, as a
function of the wind direction at Monitor 1 for all 1-hour average data
at Site #6 39
21 The ratio of the wind speed difference between Monitors 1 and 2 to the average wind
speed at the same monitors as a function of the wind direction at Monitor 1 for all
1-hour average data at Site #6. . . .' 39
22 The number of vehicles modeled versus the consecutive model hour at each site . 42
23 The intersection configuration for Site #1 (West/Chambers) used in the modeling
analysis 48
IX
-------
List of Figures (Continued)
Figure
Page
24 The intersection configuration for Site #2 (34th/8th) used in the modeling
analysis 49
25 The intersection configuration for Site #3 (65th/Broadway) used in the modeling
analysis 50
26 The intersection configuration for Site #4 (57th/7th) used in the modeling
analysis 51
27 The intersection configuration for Site #5 (34th/12th) used in the modeling
analysis 52
28 The intersection configuration for Site #6 (Battery Tunnel) used in the modeling
analysis 53
29 Average residual matched by time/location, time, and location, along with the 25-
highest unpaired values for the phase I MOBILE4.0 analysis at Site #1 86
30 Average residual matched by time/location, time, and location, along with the 25-
highest unpaired values for the phase I MOBILE4.0 analysis at Site #2 87
31 Average residual matched by time/location, time, and location, along with the 25-
highest unpaired values for the phase I MOBILE4.0 .analysis at Site #3 ,
32 Average residual matched by time/location, time, and location, along with the 25-
highest unpaired values for the phase I MOBILE4.0 analysis at Site #4
33 Average residual matched by time/location, time, and location, along with the 25-
highest unpaired values for the phase I MOBILE4.0 analysis at Site #5
88
89
90
34 Average residual matched by time/location, time, and location, along with the 25- '
highest unpaired values for the phase I MOBILE4.0 analysis at Site #6 91
35 The bias of the average versus the bias of the standard deviation for all
concentrations (paired) greater than 0.5 ppm (left-side) and the top-25 (unpaired)
concentrations (right-side) for the phase I MOBELE4.0 analysis at Site #1 93
36 The bias of the average versus the bias of the standard deviation for all
concentrations (paired) greater than 0.5 ppm (left-side) and the top-25 (unpaired)
concentrations (right-side) for the phase I MOBILE4.0 analysis at Site #2 94
-------
List of Figures (Continued)
Figure
37 The bias of the average versus the bias of the standard deviation for all
concentrations (paired) greater than 0.5 ppm (left-side) and the top-25 (unpaired)
concentrations (right-side) for the phase I MOBHJB4.0 analysis at Site #3 95
38 The bias of the average versus the bias of the standard deviation for all
concentrations (paired) greater than 0.5 ppm (left-side) and the top-25 (unpaired)
concentrations (right-side) for the phase I MOBELE4.0 analysis at Site #4 96
39 The bias of the average versus the bias of the standard deviation for all
concentrations (paired) greater than 0.5 ppm (left-side) and the top-25 (unpaired)
concentrations (right-side) for the phase I MOBHJE4.0 analysis at Site #5 97
40 The bias of the average versus the bias of the standard deviation for all
concentrations (paired) greater than 0.5 ppm (left-side) and the top-25 (unpaired)
concentrations (right-side) for the phase I MOBELE4.0 analysis at Site #6 98
41 Average residual matched by time/location, time, and location, along with the 25-
highest unpaired values for the phase n MOBILE4.1 analysis at Site #1. Also
shown for comparison are the residuals using MOBHJE4.0 emissions 116
42 Average residual matched by time/location, time, and location, along with the 25-
highest unpaired values for the phase E MOBELE4.1 analysis at Site #2. Also
shown for comparison are the residuals using MOBILE4.0 emissions 117
43 Average residual matched by time/location, time, and location, along with the 25-
highest unpaired values for the phase n MOBEJ54.1 analysis at Site #5. Also
shown for comparison are the residuals using MOBILE4.0 emissions 118
44 Scatterplots of observed versus predicted concentrations for the phase n
MOBILE4.1 analysis at Site #1 119
45 Scatterplots of observed versus predicted concentrations for the phase n
MOBELE4.1 analysis at Site #2 120
46 Scatterplots of observed versus predicted concentrations for the phase n
MOBHJE4.1 analysis at Site #5 121
47 The cumulative frequency of observed and predicted concentrations for the phase n
MOBILE4.1 analysis at Sites #1, 2, and 5 122
-------
List of Figures (Continued)
Figure
48 The 25-highest observed versus predicted concentrations for the phase n MOBILE4.1
analysis at Sites #1, 2, and 5. The solid, unmarked line is the 1:1 perfect fit ____ 124
49 The bias of the average versus the bias of the standard deviation for all
concentrations greater than 0.5 ppm for the phase n MOBILE4.1 analysis at Sites #1,
2, and 5 [[[ 125
50 The bias of the average versus the bias of the standard deviation for the 25-high
concentrations for the phase E MOBILE4.1 analysis at Sites #1, 2, and 5 ....... 126
51 The operational fractional bias (FB) with 95% confidence limits for each model
as a function of site ............................................ 132
52 The three diagnostic FB components with 95% confidence limits for each model at
60
l [[[ 133
53 The three diagnostic FB components with 95% confidence limits for each model at
Site #2 [[[ 134
54 The three diagnostic FB components with 95% confidence limits for each model at
Site #5 ....... .............................................. 135
55 The combined diagnostic FB with 95% confidence limits for each model as a
function of site ........................ . ...................... . 136
56 The combined operational and diagnostic FB with 95% confidence limits for each
model as a function of site ............. . . . . . ....... , ............. 137
57 The composite performance measure (CPM) with 95% confidence limits for each
model as a function of site ....................................... 139
58 The composite model comparison measure (CM) with 95% confidence limits using
CPM statistics ................................................ 140
59 The composite model comparison measure (CM) with 95% confidence limits using
the AFB of diagnostic category 1 ( U < 6 mph, neutral/stable ) statistics ....... 141
-------
List of Figures (Continued)
Figure
62 The AFB (DFB) with 95% confidence limits for each model pair at Site #5 144
63 The model comparison measure (MCM) with 95% confidence limits for each model
pair at Site #1 146
64 The model comparison measure (MCM) with 95% confidence limits for each model
pair at Site #2 . . . . 147
65 The model comparison measure (MCM) with 95% confidence limits for each model
pair at Site #5 14g
66 The composite model comparison measure (CM) with 95% confidence limits for each
model pair using MCM Statistics 149
A-l The cumulative frequency of observed and predicted concentrations for each model
evaluated using MOBILE4.0 emissions at Sites #1, 2, and 3 ; A-13
A-2 The cumulative frequency of observed and predicted concentrations for each model
evaluated using MOBHJE4.0 emissions at Sites #4, 5, and 6 A-14
A-3 The 25-highest observed versus predicted concentrations for each model evaluated
using MOBILE4.0 emissions at Sites #1, 2, and 3. The solid, unmarked line is
the 1:1 perfect fit A-15
A-4 The 25-highest observed versus predicted concentrations for each model evaluated
using MOBILE4.0 emissions at Sites #4, 5,and 6. The solid, unmarked line is the 1:1
perfect fit A-16
A-5 Scatterplots of observed versus predicted concentrations for each model evaluated
using MOBILE4.0 emissions at Site #1 A-17
A-6 Scatterplots of observed versus predicted concentrations for each model evaluated
using MOBELE4.0 emissions at Site #2 A-18
A-7 Scatterplots of observed versus predicted concentrations for each model evaluated
using MOBILE4.0 emissions at Site #3 A-19
A-8 Scatterplots of observed versus predicted concentrations for each model evaluated
using MOBILE4.0 emissions at Site #4 A-20
Xlll
-------
List of Figures (Continued)
Figure
A-9 Scatterplots of observed versus predicted concentrations for each model evaluated
using MOBILE4.0 emissions at Site #5 A-21
A-10 Scatterplots of observed versus predicted concentrations for each model evaluated
using MOBILE4.0 emissions at Site #6 A-22
B-l The residual or ratio of the predicted to observed concentration using the TEXIN2
model with MOBELE4.1 emissions at Site #1 plotted versus the hour of the day, wind
direction, wind speed (u), ambient temperature, Pasquill-Gifford (PG) stability
class, and traffic volume. Significant points on each box plot represent the 2nd,
16th, 50th, 84th, and 98th percentiles. The number of observations used in each box
are also labelled near the bottom as "N = #." The dashed lines represent the factor
of two lines B-l
B-2 Same as Figure B-l except for the CAL3QHC model B-2
B-3 Same as Figure B-l except for the CALINE4 model B-3
B-4 Same as Figure B-l except for the IMM model B-4
B-5 Same as Figure B-l except for the GIM model B-5
B-6 Same as Figure B-l except for the TEXIN2 model at Site #2 B-6
B-7 - Same as Figure B-l except for the CAL3QHC model at Site #2 B-7
B-8 Same as Figure B-l except for the CALINE4 model at Site #2 B-8
B-9 Same as Figure B-l except for the IMM model at Site #2 B-9
B-10 Same as Figure B-l except for the GIM model at Site #2 B-10
B-ll Same as Figure B-l except for the TEXIN2 model at Site #5 B-ll
B-12 Same as Figure B-l except for the CAL3QHC model at Site #5 B-12
B-13 Same as Figure B-l except for the CALINE4 model at Site #5 B-13
B-14 Same as Figure B-l except for the IMM model at Site #5 B-14
B-15 Same as Figure B-l except for the GIM model at Site #5 B-15
xiv
-------
ACKNOWLEDGEMENTS
Mr. Donald C. DiCristofaro of Sigma Research Corporation, Concord, Massachusetts
is the principal contributor to this document. Significant contributions were also made by
David G. Strimaitis, Robert C. Mentzer, Gary E. Moore, and Robert J. Yamartino, also of
Sigma Research Corporation. Special acknowledgement is given to Thomas N. Braverman,
the EPA Work Assignment Manager, for his diligence, assistance, and advice in using the
various modeling techniques at each of the six intersections. William Cox of the EPA is also
acknowledged for his advice in applying the EPA scoring scheme to the New York City
model results. Other people who assisted in this project include George Schewe of
Environmental Quality Management, Inc. who helped prepare the initial modeling protocol
and assisted with the intersection configurations for each of the six sites and Michael Lee of
Alice King Rosen & Fleming, Inc. who provided the New York City database and answered
many questions concerning the data.
xv
-------
-------
1.0 INTRODUCTION
1.1 Overview of the Analysis
The United States Environmental Protection Agency (EPA) is interested in updating the
guidance for modeling carbon monoxide (CO) generated by mobile sources at intersections.
The current guidance from EPA for modeling CO concentrations at roadway intersections is
to use the "Carbon Monoxide Hot Spot Guidelines" (EPA, 1978) or the "Guidelines for Air
Quality Maintenance Planning and Analysis Volume 9 (Revised): Evaluating Indirect
Sources" (EPA, 1979) for screening intersections. If the screening calculations show a
potential for exceeding the National Ambient Air Quality Standards (NAAQS) for CO, then
refined analyses are required using Worksheet 2 of Volume 9 for traffic and emissions
estimates and the CALINE3 dispersion model for concentration estimates. Both the Hot Spot
Guidelines and Volume 9 have been criticized as being outdated, inadequate, and difficult to
use. These techniques are considered outdated because (1) the major emissions components
are modal emissions factors which are based on emissions from pre-1977 vehicles; (2)
correction factors to the modal emissions model are calculated from the MOBILE1 emissions
model, which has since been updated to MOBILE4 (EPA, 1989); and (3) the traffic compo-
nent is based on the 1965 Highway Capacity Manual (HCM), which has since been updated
to 1985 (TRB, 1985). These techniques are considered inadequate because they cannot
handle overcapacity intersections. Also, these techniques are considered difficult to use
because they are in a workbook format rather than coded as a model for use on a personal
computer.
This document describes the procedures followed and results obtained in evaluating the
performance of eight modeling techniques in simulating concentrations of CO at the six
intersections monitored as part of the Route 9A Reconstruction Project in New York City.
The eight modeling techniques evaluated include:
CAL3QHC
FHWAINT
GIM
EPAINT
CALINE4
VOL9MOB4
TEXIN2
IMM
1985 Highway Capacity Manual Modified CAL3Q Model
Federal Highway Administration (FHWA) Intersection Model
Georgia Intersection Model
EPA Intersection Model
California Line Source Model
MOBELE4 Modified Volume 9 Technique
Texas Intersection Model
Intersection Midblock Model
Only two of the intersection techniques, IMM and CALINE4, include street canyon options
for modeling CO concentrations. These options have not been evaluated in this study. While
several of the intersections are located near significant buildings that may promote the
formation of circulations typically associated with street canyons, most of the emphasis is
placed on evaluating model performance at the less complex sites. The New York City
-------
database includes meteorological, CO, and traffic observations for six different intersections.
Detailed traffic information is available from the numerous videotapes available at each site.
A complete phase I model evaluation study was conducted using MOBILE4 emissions
estimates. The phase I evaluation included all eight intersection modeling techniques at all
six intersections. In late 1991, the MOBILE4.1 (EPA, 1991) emissions model, an update to
MOBILE4, was released. Thus, a phase n evaluation utilizing MOBILE4.1 was conducted
using a subset of the intersection models. As will be shown in Section 6.0 (Model Perfor-
mance Results), of the three EPA intersection models (EPAINT, VOL9MOB4, and
CAL3QHC), CAL3QHC performed best using MOBILE4. Of the two models utilizing the
FHWA advocated average speed approach rather than explicit queuing (FHWAINT and GIM),
GIM performed better. Therefore, the phase II MOBILE4.1 analysis was performed for the
following five models: CAL3QHC, GIM, IMM, TEXIN2, and CALINE4. When collecting
and compiling the New York City database, the best quality assurance procedures (analysis
and comparison of data) were followed at two of the six intersection sites, Site #1
(West/Chambers) and Site #2 (34th/8th). A uniform wind analysis (similar wind speed and
direction for different meteorological monitors at the same intersection) conducted for each
site (DiCristofaro et al., 1991) indicated that Sites #5 (34th/12th) and #1 are best in terms of
unhindered approach wind flows and wind field uniformity. Thus, the phase n MOBILE4.1
analysis was performed for the intersections at Sites #1, 2, and 5.
Two types of statistical evaluations of differences between observed and modeled CO
concentrations are performed. First, the EPA Model Evaluation Support System (MESS) is
used to calculate a standard set of performance measures and statistical estimators. Both
paired and unpaired data sets are used. Second, a scoring scheme recommended by the EPA
is used to rank the models and to evaluate the significance of the results.
1.2 Study Objectives
The ultimate objective of this study is to determine which of the eight intersection
modeling techniques most accurately simulates the highest measured CO concentrations and
whether the performance of that technique is significantly different than the other modeling
techniques. In order to achieve this objective, many other questions needed to be addressed,
such as:
• Does one model consistently display bias (i.e. overprediction or underprediction)?
• Is one model significantly better than another model (e.g. at the 95% confidence
level)?
• How well do the models reproduce the dynamic variability of the observations?
• How does the model performance vary among sites?
-------
• Are the mean errors small due to the balancing of large underpredictions with large
overpredictions?
• How does model performance vary with meteorological conditions?
• How is model performance altered by the use of observed versus worst-case
meteorological data?
1.3 Report Organization
The eight intersection modeling techniques evaluated in this study are summarized in
Section 2.0. Also included is a summary of the input data required for each modeling
technique. Section 3.0 includes a description of the six New York City intersections and the
data collected. Also included is a uniform wind analysis of the observations. The modeling
methodology used in this study including a 'description of the model input data and the
dispersion modeling techniques is presented hi Section 4.0. Section 5.0 presents a discussion
of the two types of statistical evaluations used to assess model performance. The model
performance results, including a limited number of graphs and tables for the phase I
MOBILE4.0 analysis and detailed results for the phase 31 MOBILE4.1 analysis, are presented
in Section 6.0. Also included in Section 6.0 are the scoring scheme results. Section 7.0
presents a summary of the model evaluation results and the references are listed in Section
8.0. Detailed results for the phase I MOBILE4.0 analysis are presented in Appendix A and
residual plots using the phase n MOBILE4.1 results are presented in Appendix B.
-------
-------
2.0 INTERSECTION MODELING TECHNIQUES
2.1 Overview of Modeling Techniques
The eight modeling techniques evaluated include CAL3QHC, FHWAINT, GIM,
EPAINT, CALINE4, VOL9MOB4, TEXIN2, and IMM. Six of these models are currently in
use and two are proposed for use. All of the models use the latest version of the MOBILE
emission factor model in some capacity, i.e., to estimate idle and cruise speed component
emissions or to adjust modal emissions to the scenario conditions not considered by the modal
model. The two models proposed for use, EPAINT and FHWAINT, are concatenations of
suggestions made by members of the EPA/FHWA CO Intersection Modeling Work Group.
The CAL3QHC model combines the CALINE3 dispersion model (Benson, 1979) with a
traffic algorithm to calculate queuing based on the 1985 Highway Capacity Manual (HCM)
(TRB, 1985). The GIM, TEXIN2, IMM, and CALINE4 models are procedures that have
been used over the past several years in various state programs (some of these models were
revised in the past year and these revised versions are tested hi this evaluation).
It is important to note that most of these modeling techniques are incomplete (i.e., they
do not include all necessary components for modeling CO from an intersection). The
VOL9MOB4, GIM, EHWAINT, and EPAINT models are emission and traffic movement
models only. These models use signalization, traffic volumes, and roadway capacities to
estimate traffic movements and emissions. Roadway capacities were calculated using the -
1985 HCM and emissions were calculated using MOBILE4 (Phase I Analysis) and
MOBHE4.1 (Phase n Analysis). These four modeling techniques used the CALINE3 line
source dispersion model to calculate ambient concentrations under a variety of meteorological
conditions. Two of the modeling techniques, CAL3QHC and CALINE4, are dispersion
models; CAL3QHC includes a traffic movement model and CALINE4 includes a modal
emissions model. MOBILE4 (or MOBILE4.1) modeling must be conducted separately in
order to obtain the emissions. Finally, TEXIN2 and IMM are inclusive emission, traffic
movement, and dispersion models. TEXIN2 includes CALINE3 dispersion techniques and
IMM includes HIWAY2 dispersion calculations. These two models also directly incorporate
the MOBILE4 model so that emissions estimates are calculated internally. These two models
have been revised to use MOBILE4.1 for the phase n modeling analysis.
For the phase I analysis using MOBILE4, the model combinations required are
summarized below:
Model No. 1
HCM + MOBILE4 + EPAINT + CALINE3
Model No. 5
MOBILE4 + CAL3QHC
-------
Model No. 2
HCM + MOBELE4 + FHWAINT + CALINE3
Model No. 3
HCM -t- MOBILE4 + VOL9MOB4 + CALINE3
Model No. 4
HCM + MOBILE4 + GIM + CALINE3
Model No. 6
MOBILE4 + CALINE4
Model No. 7
TEXIN2 (includes MOBELE4)
Model No. 8
IMM (includes MOBILE4)
For the phase II analysis using MOBILE4.1 emissions estimates, the model combinations
required are summarized below:
Model No. 1
HCM + MOBILE4.1 + GIM + CALINE3
Model No. 2
HCM + MOBILE4.1 + CAL3QHC (Version 2.0)
Model No. 3
MOBILE4.1 + CALINE4
Model No. 4
TEXIN2 (includes MOBHJ24.1)
Model No. 5
IMM (includes MOBELE4.1)
Note that a revised version of CAL3QHC (Version 2.0) was used for the second-phase
modeling analysis so that differences in performance between phase I and phase n are not
solely due to replacing-MOBILE4.1 with MOBILE4. Version 2.0 of CAL3QHC allows the
user to input the saturation flow rate, signal type, and arrival rate. Other changes include
modification of the queue delay and queue length calculations.
2.2 Model Summaries
Primary differences among the eight modeling techniques are due to emission, traffic,
and roadway characterizations rather than dispersion modeling methods. Each of the eight
models evaluated except IMM use a form of the CALINE3 model for dispersion estimates.
The IMM model uses HIWAY2 dispersion modeling techniques. Each model is briefly
described below along with the additional model components required to estimate ambient CO
levels. Table 1 describes the input data needed for each model.
-------
I
ffl
OS
as
H
CJ
1
5
I
C/3
.
3
.1-
2
"3
Ofl
1
§
1
12
£
J
TS
•3,
3
•3
"o
-J
a
•5
O"
4J
as
f
•3
^
-
S
1
,g
|s
3 0
< 2
U *
oS^p"
Sis!
d??l
§._. en
1 = ^5
O •)••*• •(.
C3 3! en
o 3 a
? 2 3 R
O — S **
>~ -c1^
1 11
g o o <
s ~ •§ ^
I3
-< O O -<
&? * *
"u
•3
Intersection M
Inputs
I Roadway Configuration:
X
X
n
0
1
J,
X
•§•
X
i
Midblock- 10 Midblock Disi
X
X
X
X
X
X
•g
.2
J
IM
O
Z
X
X
X
u
u
u
u
1 Number of Links
's
X
X
1?
S
X
s"
X
f
> §
s s
If
85
!H
3 s|
9^9
2 ^"^
2 gcS
•S .3 .3
111
= 33
Sxo
-------
1
c
I
S
I
1
|
._§
3
S
i
df
•8
£
1
*S
sr
as
3.
"""
j
^
i
CAUNE4
+MOB1LB4
0 T S3"
cy "™* L'J' s
2 o § "^
S?5l
5 3
S 3 §
^ O O <•
§7??
-W)
X
*a
|| Observed Queue Length (optic
•w*
X
Observed Delay (optional)
is
S 31
II
§•!
"2 ""
a -3
3> S
U w
•M
l
s a q
.3 .S
11
2XC
-------
I
S
«
S
8
•if
i
1
ao
01
•3
for Eacli
1
f
1
.5
•3
•5
J
2
S
^
g
TT
S £ij
§ 3
3 9
S?
S + ? :>
11
0 * * *
5_
||§1
>~ + +
i si
Illl
- S|
**• — § ""•
S ~ +• +•
1
1 Dispersion/Meteorology:
X
£
2
0
a-
X
£
2
2
£
X
£
X
£•
X
£
°_
X
Temperature Per Hour
f
X
X
f
X
X
1
CJ
1
CJ
f
CJ
*-^
,5.
CJ
Wind Speed
1
X
1?
2,
X
1
X
I
X
"3
u
3.
CJ
1
CJ
1?
o
"M
2,
CJ
Wind Direction
X
X
X
X
CJ
CJ
CJ
CJ
g
Aunospheric Stability Cla
?
X
£,
X
?
X
1
X
1
CJ
1
CJ
CJ
,§
o
.2?
oa
^c
2
X
i
X
X
1
CJ
J
CJ
J
CJ
J~
CJ
g
s
CO Background Concentr
'?
1
X
1
X
1
X
?
0
1
CJ
CJ
-;.
£
CJ
Surface Roughness
X
I
X
j
CJ
1
CJ
1
CJ
•?
c
U
4)
|
X
!
X
j
CJ
1
CJ
1
CJ
•S}
JJ,
CJ
X'
1
>
1
iJ
CO
1
X
1
X
|
o
1
CJ
CJ
•?
CJ
Deposition Velocity
S
X
I Altitude of Site
X
§
M
•d
•^
Standard Deviation of Wii
a.
at
X
X
X
X
CJ
CJ
CJ
CJ
|| Number of Receptors
X
A
X
,pf
X
-?
X
ft
CJ
'a
S
CJ
S
CJ
¥
•^
a
I
•a.
•3
"s
X
X
s
«
X
•?
s
X
s
u
J
CJ
I
CJ
s
ft
a
i
Hcigia of Receptors
1
II
II
f-%
•3
— »>
•r|
H
"• »
•o •;!
i "3
.2> S
u w
.3
II i
3 la
ss^
ifS
•S .3 .3
1 'S'S
= 33
2 x cj
-------
1
c
•3
a
3
P3
3
30
J
|
1
I
•8
,§
f"
as
M
1
a
*
g
P
T§
3?
0 ^
I'll"!
s??l
3 S
„ g S
s go 5
s * *"
1 ||
>~ f •*•
Ul
— s S S
S~ f »
9,
•< o o ^
&~ * *
Inlcncclion Model
Iltpuls
Traffic/Slunall/allon:
1"
|
1
X
}
X
X
I
s
•S
,§
X
S"
f
c
*wT
•X
1
X.
S
1
X
S
| Free Flow Speed Per Approach
vt
X
•2"
S
Jf
X
3
X
3
X
3
X
Green Time Per Approach
3
X
X
3
X
3
X
X
3
X
3
X
| Toial Cycle Time
"j?
J
X
1
X
^
•&
X
I
X
*c*
*§
1
o
tf
I
X
y"
•S.
X
y"
j
v^
°"
|| Approach Traffic Volume Per Hour
^
X
I Running Time Per Mile
*J?
J
1
X
*4J*
I
S
X
f
X
o
•g
i
X
Toul Capacity Service Volume/Green
Tune
X
X
|| Number of Signal Phases
X
•a
|| Fraction of vehicles turning left per lii
X
J
I
i
00
I
o
£
X
| Left-Turn Signalizalion Type
s
X
e
o
1
u
3
'c
,..
X
1 Red Tune
,-.
X
Single Vehicle Clearance Time
3
^S'
X
| Deceleralion/Acceleration Speeds
X
|| Duculerauon/Accclcralion Time
X
|| Average No. Veh. per Cycle per Lane
X
o
J
| Avg. No. Veh. Delayed per Cycle Per
X
|| Depanure Traffic Volume per Hour
3
X
1 Vehicle Idle Time at Slopline
X
|| Vellide Idle Time at End of Queue
i- 2
8-2
11
3 s|
3 8«
§!"<
•« 3 o
•S .S .3
111
= 33
2 X O
10
-------
2.2.1 EPA Intersection (EPAINT) Model
An EPA-proposed traffic and queuing technique for estimating CO emissions from
approaches to intersections is referred to as the EPAINT (EPA Intersection) model (PEI,
1988). This technique falls into the class of a mobile source model that estimates vehicular
emissions and queuing at an intersection. The EPAINT model requires the external use of
both MOBILE4 and the 1985 Highway Capacity Manual (HCM). The technique explicitly
treats vehicles that are delayed at an intersection. In the EPAINT technique, vehicle
movements are separated into a free-flow component and a delayed or queued component.
The combination of the two overlapping roadway segments yields the EPAINT estimate of
CO emissions and the distances over which they apply at each approach to the intersection.
In the EPAINT model, the arterial speed is adjusted for vehicle volumes, roadway
capacity, and any other roadside frictions (i.e., driveways, businesses, and cross streets) that
reduce capacity. This speed is used to estimate an adjusted free-flow speed on the roadway
segment. The HCM Chapter 11 technique for estimating arterial speed was modified for use
in EPAINT by excluding the effects of delay from the average arterial speed calculation. The
composite CO emissions for the segment are calculated via MOBILE4 by using the modified
arterial speed and other ambient and operating conditions.
Excess emissions due to delay are calculated in EPAINT by using an adjusted idle
emission factor from MOBTLF.4, the total approach delay time per vehicle, and the volume of
traffic on the approach. For this evaluation, the MOBILE4 idle emissions were adjusted by
using the ratio of a scenario composite emission to a base-case composite emission at 2.5
mph. The idle emissions are applied over an excess emissions distance calculated by queuing
techniques given in the Institute of Traffic Engineering (ITE) Handbook. This model was not
tested with the MOBILE4.1 emissions model..
In order to facilitate the use of the EPAINT model and to make the model consistent
with current modeling guidelines, the following changes were made to the computer
algorithm:
• Code was converted from an interactive mode to batch mode;
• More than one link at a time may be modeled;
• The idle and base idle vehicle speed was changed from 5.0 to 2.5 mph;
• Allow the vehicle mix, annual mileage accumulation rates, registration distribution,
refueling emissions options, RVP (Reid Vapor Pressure), I/M (Inspection/
Maintenance) and ATP (Anti-Tampering Program) parameters to be input to the
model, rather than fixed as constants within the code; and
11
-------
• Allow the thermal states for idle and scenario conditions to be input to the model,
rather than fixed as constants within the code.
The adjusted free-flow and queuing emissions estimated by EPAINT are input to the
CA1INE3 dispersion model. Input link information to CALINE3 is tailored to fit the
EPAINT-generated queue lengths for each scenario. The EPAINT results are formatted to the
gram-per-vehicle mile input units required by CALINE3.
2.2.2 FHWA Intersection (FHWAINT) Model
While the EPAINT model divides the vehicles into a free-flow component and a delayed
component, the FHWA-proposed technique, known as the FHWAINT (FHWA Intersection)
model (PEI, 1988), calculates an adjusted vehicle speed and related composite CO emission
on the approach to accommodate vehicle delay. This technique estimates the emissions over
a length of user-selected roadway (segment) on the basis of the volume to capacity (V/C)
ratio. The V/C ratio is used to determine the average speed of a vehicle over the whole
segment, which includes the effects of the delay of the vehicle at an intersection. FHWAINT
includes the current MOBILE4 model to estimate the composite CO emissions at the adjusted
vehicle speed. The HCM model is used to calculate the roadway capacities. The use of
FHWAINT for V/C ratios greater than 1.0 is not recommended by FHWA. The resulting
emissions represent a composite free-flow and queuing link with the overall cycle being
represented by a lower vehicle speed (and subsequent higher CO emissions).
Changes to the FHWAINT computer algorithm, similar to the changes discussed above
for the EPAINT model, were made in order to facilitate the use of the FHWAINT model and
to make the model consistent with the other models being evaluated. For example, the idle
and base idle vehicle-speed was changed from 5.0 to 2.5 mph. Also, the vehicle mi*, annual
mileage accumulation rates, registration distribution, refueling emissions options, RVP, VM,
ATP, and thermal states were input to the model rather than fixed as constants.
The EHWAINT-calculated composite emissions are input to the CALINE3 dispersion
model. The FHWAINT technique assumes that the free-flow and queuing emissions have
been accounted for by the adjusted speeds of the approaches; thus, no queue links are
included in the dispersion modeling. All other CALINE3 components of the analysis are
identical to routine CALINE3 applications. This model was not tested with the MOBILE4.1
emissions model.
2.2.3 VOLUME9/MOBILE4 (VOL9MOB4)
The previous versions of the VOLUME9 model used the MOBILE1 model for adjusting
emissions, the VOLUME9 Appendix B capacity analyses (based on the 1965 Highway
Capacity Manual analysis), and the fflWAY model. In keeping with current
12
-------
recommendations but including the basic techniques in VOLUME9, the MOBHJE4 emissions,
1985 Highway Capacity Manual calculations for roadway capacity, and the CALINE3 model
were used to supplement the VOLUME9 analysis. This is referred to as the VOLUME9-
MOBILE4 technique or VOL9MOB4. For delay, queue length, and excess emission
calculations, the procedures previously used in Volume 9 have been maintained and follow
the 1965 Webster Techniques. The VOLUME9 (EPA, 1979) Worksheet 2 calculations for
determining emissions and traffic at an intersection have been computerized to allow quicker
calculations and direct access to the MOBILE4 model. Worksheet 2 specifically addresses
the calculation of excess emissions and the length of roadway over which they take place.
The HCM model is used to calculate the roadway capacities.
The overall emissions in VOL9MOB4 consist of free-flow, acceleration, and deceleration
emissions, which are estimated based on the Modal model (Kunselman, 1974) for a 1977 base
case. The idling emissions are based on MOBILE4 and are tabulated in the same mass/
vehicle/ distance units as free-flow emissions. Estimates of these emissions are based on the
number of vehicles, the proportion of vehicles that stop, and the average vehicle delay rime.
The excess emission segment length (resulting from queuing, acceleration, and
deceleration) is the greater length arrived at by two separate techniques. The first is the
length needed for a vehicle to decelerate from cruise speed to a stop and then accelerate back
to cruise speed. The second length is calculated as a function of the number of vehicles that
stop and an average vehicle length (8 m). The greater of the acceleration/deceleration length
or queuing length is used for excess emissions. The free-flow roadway length is user-
specified. The results of this procedure are input to the CALINE3 model for all dispersion
estimates. Separate free-flow and excess emission links are modeled with CALJNE3. As
shown in Table 1, the input requirements for VOL9MOB4 are similar to EPAINT and
FHWAINT. This model was not tested with the MOBILE4.1 emissions model.
Changes to the VOL9MOB4 computer algorithm, similar to the changes discussed above
for the EPAINT model, were made in order to facilitate the use of the VOL9MOB4 model
and to make the model consistent with the other models being evaluated. For example, the
idle and base idle vehicle speed was changed from 5.0 to 2.5 mph. Also, the vehicle mix,
annual mileage accumulation rates, registration distribution, refueling emissions options, RVP,
I/M, ATP, and thermal states were input to the model rather than fixed as constants.
2.2.4 Georgia Intersection Model (GIM)
The Georgia Intersection Model (GIM) technique calculates traffic flow and emissions
from intersections, based on a modified U.S. EPA VOLUME9 approach (EMI, 1985). This
model was designed by the Georgia Department of Transportation (GDOT) to handle under-
capacity, at-capacity, and over-capacity scenarios. The output of GIM is designed to be input
directly into an air dispersion model; in this case, the CALINE3 model is used. The HCM
model is used to calculate the roadway capacities. MOBELE4 model emission estimates are
13
-------
necessary for GIM use. This modeling technique was also evaluated using the MOBELE4.1
emissions model.
The GIM model calculates an effective excess emissions length of roadway from the
point at which vehicles begin to decelerate upstream of an intersection. This distance
includes the length of road where cars slow down and where they queue in the upstream
direction. Over this length, vehicle speeds are reduced to account for delays caused by
vehicles slowing and stopping at the intersection. The GIM model calculates the average
speed over this distance (thereby accounting for the delay) and estimates the average CO
emission rate using MOBILE4 emissions factors for vehicles traversing the affected length.
Using this approach, the use of modal emission factors is not necessary. The GIM output
defines finite line source segments with their associated CO emission rates, which are used as
input for the CALINE3 dispersion model. Roadway segments are not separated into idle and
free-flow emission components. The user, however, must generate emissions using
MOBILE4 for those portions of the roadway that are not associated with the effective excess
emission lengths, i.e., any free flow extensions beyond the GIM-generated links that complete
the characterization of the approach and departure links.
2.2.5 CAL3QHC
The CAL3QHC (EPA, 1992) model was developed by EPA Regional Offices I and IV to
calculate CO concentrations at intersections. The CAL3QHC model is a hybrid of the
CALINE3 line source dispersion model and an algorithm for estimating vehicular queue
lengths at signalized intersections. No modal emissions due to acceleration or deceleration
are included in this model explicitly; instead, they are included implicitly in the Federal test
procedure cycles in the MOBILE4 calculations. The models and techniques used in
CAL3QHC are 1) the MOBILE4 model emissions, which are estimated separately from
CAL3QHC for free-flow and idling (adjusted to scenario conditions); 2) the delay procedures
of the 1985 Highway Capacity Manual (and associated queuing); and 3) the CALINE3
dispersion model. The latter two components are included directly in the CAL3QHC model.
In the CAL3QHC model the excess emissions or linear source strength for stopping
vehicles are based on the red time, the number of lanes, and adjusted MOBHLE4 idle
emissions (adjusted for cold/hot starts, temperature, vehicle mix, etc.). The emission rate is
then set equal to a constant (100 g/veh-mi) and the number of vehicles that represent the
linear source strength is calculated. The queue length is calculated on the basis of traffic
volume, signal cycle time, red time, clearance lost time, and a vehicle length of 6 meters.
The queue represents only the idling emissions. Free-flow emissions are handled separately
by another, overlapping roadway segment length. These links are then used in the CALINE3
portion of the CAL3QHG model with associated roadway and receptor geometry and
meteorological conditions. •
14
-------
A revised version of CAL3QHC (Version 2.0) was tested with the MOBILE4.1
emissions model. The revised version of the model addressed comments received in response
to the Fifth Air Quality Conference. The objectives of the modifications to the CAL3QHC
model were to: 1) give the user more freedom of choice (in terms of capacity determination,
signal type, and arrival rate); 2) base the choices on recommendations from the 1985
Highway Capacity Manual (HCM); and 3) keep the same input/output formats from the
original version. These modifications affect the calculation of the V/C ratios and queue
length.
The three new input variables that can be specified or set by default are:
1) Saturation Flow Rate or Hourly Capacity per Lane
The saturation flow rate is determined by the user depending on the characteristics
and operation of the intersection. If no input value is used, the program assumes
1600 vehicles per hour (vph) as representative of an urban intersection.
2) Signal Type
The signal type may be set to either pretimed, actuated, or semiactuated. The
default value is pretimed.
3) Arrival Rate
The arrival rate may be set to either worst progression (dense platoon at beginning
of red), below average progression (dense platoon during middle of red), average
progression (random arrivals), above average progression (dense platoon during
middle of green), and best progression (dense platoon at beginning of green). The
] model assumes random arrivals as the default
The signal type and arrival rate are used by CAL3QHC (Version 2.0) to calculate the
progression adjustment factor that will affect the delay calculation. Two other internal
modifications to the model include adjusting the queue delay and the queue length. The delay
for the queue calculation is based on the total approach delay in the new version rather than
the stopped delay as used in the original version. In addition, the third term of the original
Webster formula for the queue length calculation has been reinstated. This will only have an
effect on low V/C ratios.
2.2.6 CALINE4
The CALINE4 model (Benson, 1989) is a line source air quality model developed by the
California Department of Transportation as an update to the previous CALINE3 model. The
15
-------
CALINE4 model includes the capability of handling modal modeling components including
delay at an intersection by treating individual vehicle delay, acceleration, deceleration, and
free-flow. Cumulative modal emission profiles are constructed for each link based on speed,
acceleration/deceleration rates, idle (delay time), and traffic volumes on each link. The
CALINE4 model includes modal emissions and dispersion components, but does not include a
traffic model component The emissions from stopped vehicles are based on an emissions
profile that is generated from an assumed constant annual rate of vehicles. The cumulative
emission profile is then generated as a function of the time spent by each vehicle in each
mode. The MOBTLF.4 emissions are required for both a specified set of vehicle operating
conditions for a composite emission factor as well as a scenario adjusted idle factor. This
model was tested with both MOBILE4 and MOBELE4.1 emissions.
2.2.7 TEXIN2
The TEXIN2 model (Bullin et al., 1990) was developed by the Texas Transportation
Institute (TIT). The MOBILE4 model is incorporated directly into TEXIN2 such that the user
specifies the vehicle speed, year of analysis, temperature, and other operating conditions and
scenario specifications for the overall model. No idle adjustments for ambient temperature
and hot/cold start conditions are made. A revised version of the TEXIN2 model with
MOBHJE4.1 emissions was also tested. For this version of the model, the idle adjustments
were automatically made by MOBILE4.1. .
Traffic is handled by the TEXIN2 model by using the Critical Movement Analysis
(CMA) Operations and Design Technique. The CMA technique treats the intersection as a
unit and considers conflicting movements- that must be accommodated, The resulting traffic
volumes are used together with cruise and excess emissions to form the source terms for each
link. Excess emissions are calculated as a function pf two vehicle operating modes: 1) vehi-
cles slowing but not stopping; and 2) vehicles that stop and idle. For the first component, a
composite emission rate for one-half the link free-flow speed is used along with approach
delay and time in the queue. For stopping vehicles, the unadjusted MOBILE4 idle factor is
used along with adjusted modal emissions factors for acceleration and deceleration.
TEXIN2 treats each leg of an intersection as a link and individual lanes are not
considered in the model. TEXIN2 also includes traffic delay calculations as well as the
CALTNE3 dispersion modeling component. One adjustment made by TEXM2 to the
CALINE3 model is the application of a factor for low wind speed cases.
2.2.8 Intersection Midblock Model (IMM)
The Intersection Midblock Model (IMM) (NYDOT, 1982) was originally developed by
GCA Corporation under contract to the EPA in 1978 as part of the Carbon Monoxide Hot
Spot Modeling Guidelines (EPA, 1978), but was later revised and updated by the New York
16
-------
Department of Transportation. The IMM model combines the use of various components
required for a highway, street, or intersection analysis into one computerized technique. The
IMM model was based originally on the VOLUME 9 techniques using an excess emissions
approach including modal emissions. The model includes modal emission calculations for
delayed vehicles at an intersection which are calculated on the basis of vehicle stopping and
starting movements. The IMM model has been updated with MOBILE4 for making emission
estimates and adjustments of modal emissions. The model is capable of estimating CO
concentrations at receptors near intersections, at midblock locations, and in street canyons.
The IMM model will accept data for two intersections with a maximum of four phases per
intersection. The HIWAY2 model for line source dispersion calculations is used for all
atmospheric transport and dispersion analyses. A revised version of the IMM model with
MOBTLFA1 emissions was also tested.
Emissions from accelerating/decelerating vehicles and idling vehicles are assigned to
pseudolinks which are lengths along the link where the emissions emanate on average. The
acceleration/deceleration rates are used to compute the pseudolinks. Traffic signal
characteristics and capacity service volumes are used to calculate the queue length and delay
time which then determine the idle emissions. The cruise and acceleration/deceleration
emissions are calculated by use of the EPA Modal Analysis Model (Kunselman, 1974), which
has been incorporated in IMM.
17
-------
-------
3.0 THE NEW YORK CITY DATABASE
3.1 Description of the Six New York Intersections
A major air quality monitoring study was conducted in 1989-1990 in response to the
proposed reconstruction of a portion of Route 9A in New York City. The reconstruction is
proposed for the southernmost five miles of the roadway from Battery Place to West 59th
Street. As part of the monitoring project, meteorological and CO air quality data were
collected at two background sites and six different intersections. These sites are all located in
midtown or lower Manhattan, and are shown in Figure 1. Three of the sites (Site #1
West/Chambers; Site #5 12th/34th; and Site #6 Brooklyn Battery Tunnel) are on the Route
9A Right-of-Way. Layouts that identify locations of the meteorological monitors (labeled as
Ml, M2, etc.), the CO monitors (labeled as PI, P2, etc.), and nearby buildings at each
intersection are shown in Figures 2 through 7,
Two of the six intersections are "unobstructed" sites with relatively few nearby buildings
or structures. Site #1 is located at the intersection of West Street (Route 9A) and Chambers
Street in the vicinity of Battery Park City along the Hudson River. The site is relatively open
with a parking lot on the southeast side of the intersection and low buildings (5 to 30 m)
extending from the east southeast to the north northeast of the intersection. Site #5 lies along
the Hudson River at West 34th Street and 12th Avenue (Route 9A) adjacent to the Jacob
Javits Exposition Center. There is virtually unobstructed flow over the Hudson River from
the south southwest to the north. There are low buildings (one to three stories) to the east
and south. Site #5, along with Site #1, represent the best intersections with respect to
unobstructed flows.
Two of the six intersections are, street-canyon sites. Site #2 is a midtown intersection at
34th Street and 8th Avenue. The intersection is one block north of Madison Square Garden
and the General Post Office Building. There are skyscrapers up to approximately 100 to
150 m in height on all sides of the intersection. Site #4 is a midtown intersection at West
57th Street and 7th Avenue at Carnegie Hall. This is also a street-canyon setting with tall
buildings (up to 70 stories) on all sides.
The final two intersections analyzed are complicated by a number of factors. Site #3 lies
at the convergence of Columbus Avenue, Broadway, and West 65th Street in the vicinity of
Philharmonic Hall and the Lincoln Center. There are five and six-story buildings on all sides,
although the intersection center is relatively open. Of the six intersections analyzed, this site
has the most complicated configuration (e.g., adjacent traffic lights and intersections). Site #6
is at the intersection of the Brooklyn Battery Tunnel with West Street (Route 9A). Data were
collected at two sites (6A and 6B) in the vicinity of the tunnel. Because traffic data are not
available from Site #6B, only Site #6A was analyzed. There are tall buildings from eight to
forty-four stories on all sides of the intersection. This site is complicated by an overhang
associated with the Port of New York Authority Building under which traffic departs from the
tunnel.
19
-------
Site #3
Site #4
Site #2
Site #5
Post Office
Background Station
Battery Park
Background Station
MANHATTAN
BMMMM MAM AVTOMMU MUTES
MATn* !••
Site
Figure 1. Locations of the six intersections and two background stations in the New York
City database.
20
-------
CDN
M2
Battery
Park"
Citv
j
BACKGROUND
SITE '
/ I
Manhattan
Community
Center
50
150
200 FEET
Figure 2. Site #1, West/Chambers, location map. The Battery Park background site is also
shown.
21
-------
W.35THST.
3
Storv
9
Storv
43
Storv
15-23 Stories
4-6 Stories
P6
P7.
W.34THST. • P5,M3
4
Storv
W.33RDST.
5
Story
BACKGROUND
»SJTE
P4
22
Storv
t
PI
P2,M1
P3
P8, M2
SITE MT-1
1
Storv
57
Storv
Madison
Square
Garden
soo
SCU£
Figure 3. Site #2, 34th/8th, location map. The Post Office background site is also shown.
22
-------
5 Storv
2 Storv
JLTLLIARD SCHOOL
OFML'SIC
5 Storv
WEST 65TH STREET
M2.P7'
PHILHARMONIC HALL
5 Storv
50 WO 1SOFHET
SCALE
•
P2
36 Story
5 Ston-
• H4
Figure 4. Site #3, 65th/Broadway, location map.
23
-------
22-35 Storv
£
&
ft*
3-4 Story
u
•1
•HEOJ
3B
12 Story
12 Story
ORNE
HOTEL
P8,M1
4 Story
IStorv
14 Story
14-45 Story
0 50 '00 ISO 200 2SOFSST
I • ' ~~1
SCALH
Figure 5. Site #4, 57th/7th, location map.
22Storv
WEST 58TH STREET
12-D Story
7 Story
9 Stan-
P7
6 Story
13 Storv
6 Ston-
P4.
PI
P3,M2
WEST 57TH STREET
CARNEGIE
HALL
60
Story
£•
-------
P6
M2,P7«
P5
P4
2 Storv
3Storv
-*P3.
WEST 34TH STREET
i tf
P8l
• P2
D
D
IStorv
WEST 33RD STREET
0 SO WO ISO 200 2SOFST
SCALE
Figure 6. Site #5, 34th/12th, location map.
25
-------
N
•MStorv
PI
W.THAMES STREET
LsJ
P4
t
8 Story
PORTOFNEWTORK
AUTHORITY
P3
MORRIS STREET
H Story
35 Story
Figure 7. Site #6A, Battery Tunnel, location map.
26
-------
There are two sites which collected background data. One background site is located on
the Battery Park City landfill near Sites #1 and 6. The second background site is located on
top of the General Post Office Building across the street from Madison Square Garden one
block south of Site #2 and near Sites #3, 4, and 5.
3.2 Description of the Data Collected
The configuration, operation, data processing, and quality assurance/quality control
practices for this program conformed, as close as possible, to the provisions of EPA's
Ambient Monitoring Guidelines for Prevention of Significant Deterioration (PSD) (EPA,
1987a). The meteorological data collected at each intersection include wind direction, wind
speed, temperature, and sigma theta (GQ). The background site at Battery Park provides both
meteorological and CO measurements, but only CO measurements from the rooftop are
available at the Post Office site. A summary of the available meteorological and CO
monitoring data is given in Table 2. The meteorological measurements were taken at a height
of 10 m ± 1 m. The CO probe heights for each monitor and site are given in Table 3.
Further details concerning the monitoring program are given in ENSR (1988).
In order to obtain detailed information concerning the traffic characteristics, a series of
video cameras were used to film the traffic at each site. Three months of continuous traffic
data were collected at each site producing approximately 13,000 hours of video recordings. A
limited number of videotaped hours were examined for the Route 9A Study in order to obtain
detailed information about the local traffic (see Table 2). The traffic data were concurrent
with the observed CO data. The examined traffic data are comprised of the top 50 hours of
CO concentrations observed for each of three months at Sites #1 and 2 and the top 25 hours
observed for each of three months at the remaining sites. Some sites listed in Table 2 have
less than the maximum 150 or 75 hours over the entire three-month period, because we have
used only those hours for which all monitors at a site had observed CO concentrations greater
than 3 ppm.
Traffic-related variables that are available for each selected hour are listed in Table 4.
All traffic data were obtained from videotapes except for the acceleration/deceleration rates
and the cruise speed. The acceleration/deceleration rates and cruise speeds were obtained
through the use of a vehicle outfitted with a travel-log machine that recorded instantaneous
speed versus time while traveling. Cruise speeds were taken directly from the strip charts
created in this way; acceleration/deceleration rates were determined from the slope of the
lines on the strip charts (Conway and Zamurs, 1991). The modified average speed is the total
travel time less the average stop delay time on the link.
The traffic data at each site are reported for a number of intersection segments or links.
For example, at Site #1, the traffic data are reported for 17 different links of the West/
Chambers intersection (e.g., westbound, northbound, other nearby intersections). Other data
27
-------
TABLE 2
SUMMARY OF THE AVAILABLE ROUTE 9A
RECONSTRUCTION PROJECT MONITORING DATA
Site Location
No.
1 West/Chambers
2 34th/8th
3 65th/Broadway
4 57th/7th
5 34th/12th
6A Battery Tunnel
6B Battery Tunnel
Bkgrd Battery Park
Bkgrd Post Office •
Collection
Period
2/89 - 5/89
5/89 - 11/89
11/89 - 1/90
11/89 - 1/90
8/89 - 12/89
11/89 - 3/90
11/89 - 3/90
1/89 - 4/90
5/89 - 4/90
# of Met
Towers
2
3
2
*2
2
1
1
1
0
#of CO
Monitors
8
8
8
6
8
4
4
1
1
# of Examined
Traffic Hours
142
143
66
74
75
75
0
-
28
-------
TABLES
CO PROBE HEIGHTS (FEET) FOR EACH MONITOR
Site
No.
1
2
3
4
5
6A
PI
10.50
9.84
11.00
11.00
9.84
9.84
P2
9.50
9.84
11.00
11.00
9.84
9.84
P3
10.00
9.84
11.00
11.00
9.84
9.84
CO Monitor
P4 P5
10.00
9.84
11.00
11.00
9.84
9.84
9.75
9.84
11.00
11.00
9.84
P6
10.00
9.84
11.00
11.00
9.84
P7
9.00
9.84
11.00
11.00
9.84
P8
10.0
9.84
11.00
11.00
9.84
29
-------
TABLE 4
SUMMARY OF AVAILABLE HOURLY TRAFFIC DATA
FOR EACH INTERSECTION SEGMENT
Vehicle Mix (Fraction)
Automobiles
Fleet Medallion NYC Taxis
Non-Fleet Medallion NYC Taxis
Non-Medallion NYC Taxis
Light-Duty Trucks
Heavy-Duty Gas Trucks
Heavy-Duty Diesel Trucks
Traffic Data
Volume (vehicles per hour)
Average Speed (mph)
Stopped Delay (sec)
"Modified" Average Speed (mph)
Queued Vehicles (vehicles per lane)
Cruise Speed on Block (mph)
Cruise Speed on Downstream Block (mph)
Number of Lanes
Cycle Time (sec)
Acceleration Rate (mph/sec)
Deceleration Rate (mph/sec)
30
-------
available on an average basis for each intersection include the thermal state conditions which
were obtained from field interviews and the average turn movements.
3.3 Analysis of the Observations
3.3.1 Analysis of the Wind Structure
In order to evaluate the appropriateness of the collected data for the model evaluation
study, a series of data distribution analyses were prepared for each intersection site using all
available hourly averaged data. It is preferred that the wind field at an intersection be
uniform for the intersection modeling techniques. Thus, comparisons were made of wind
direction, sigma theta, and wind speed at the different meteorological monitors at each
intersection site. The entire analysis is discussed in detail in DiCristofaro et al. (1991). For
this report, two different types of data plots using all available hourly-averaged data over the
entire collection period at each site are presented:
Plot Type 1
cos(0) vs. wind direction (WD) where the angle 9 is the difference in wind direction
between two different monitors
Values of cos (9) equal to one indicate perfect wind direction alignment between the two
monitors, values approaching 0.0 indicate measurements that differ by 90°, and values
approaching -1.0 indicate 180° difference in flow that may be associated with street canyon
rotors. For this data analysis, spatially uniform wind fields are arbitrarily defined by those
cases for which cos(9) > 0.85, or the wind direction measurements are within 32° of each
other.
Plot Type 2
IWS1 — WS21
-i — !• vs. wind direction where WS1 is the measured wind speed at Meteorolo-
WS
gical Monitor #1, WS2 is the measured wind speed at Meteorological Monitor #2, and
WS is measured average wind speed.
IWS1 - WS21
Values of -! — !• equal to 0.0 indicate perfect agreement between wind speed
WS
measurements. For this data analysis, uniform wind speed fields are arbitrarily defined as
those hours for which \WS1-WS2[ ^ Q4
WS
. 31
-------
Plot types 1 and 2 for Site #1 are shown in Figures 8 and 9. In general, Site #1 is
relatively open with a parking lot on the southeast side and low buildings to the northeast.
As shown in Figure 8, the majority of the values of cos(0) are greater than 0.85. The gaps or
sparsity of Monitor 1 wind direction data from 0 to 75° and 130 to 190° indicate the blocking
influence of nearby buildings (see Subsection 3.1). The wind speed difference (plot type 2)
plotted as a function of the Monitor 1 wind directions are shown in Figure 9. The majority
IWS1 - WS21
of the J == L data are less than 0.4 which is indicative of uniform winds
WS
Site #2 is located near Perm Plaza in an area of very tall buildings. Meteorological
measurements were made at three different locations, two on West 34th Street and one on 8th
Avenue. Plot types 1 and 2 for Site #2 are shown in Figures 10 through 13. The wind
direction measurements from Meteorological Monitors #1 and 2 are compared in Figure 10.
As shown in Figure 3, Meteorological Monitor #1 is located on 8th Avenue and Monitor #2 is
located on 34th Street The data indicate the presence of complex flows including street
canyon rotors and strongly channeled flows. As shown in Figure 12, non-uniformity in wind
directions is also found using Meteorological Monitors #2 and 3 which are both on 34th
Street. Although, the uniformity in wind direction is poor at Site #2, the uniformity in wind
speed is good (see Figures 11 and 13).
Plot types 1 and 2 for Site #3 are shown in Figures 14 and 15. Figure 14 indicates large
differences in 9 with large gaps at both wind direction monitors. The large wind direction
gap is due to the presence of Philharmonic Hall and the Julliard School of Music which lie to
the southwest and west of Monitor #1. The. wind speed differences shown hi Figure 15 also
indicate non-uniform wind fields.
Site #4 is located at West 57th Street and 7th Avenue near Carnegie Hall. A variety of
building heights from 1 to 70 stories are located in the vicinity of the intersection. For Site
#4, plot types 1 and 2 are shown in Figures 16 and 17. In Figure 16, the values of cos(9)
plotted versus the wind direction at Meteorological Monitor #1 indicate uniform wind
directions clustered from 90 to 140° and from 280 to 330°. Figure 17 indicates that this site
is not uniform in terms of wind speeds.
Site #5 is located near the Jacob Javits Exposition Center along the docks on the Hudson
River. There is a wide fetch with little or no building influences from 200 through 30°. The
buildings to the east and south are three stories or less. Plot types 1 and 2 for Site #5 are
shown in Figures 18 and 19. The majority of the cos(9) data plotted in Figure 18 approach
1.0 for almost all wind directions. There is some scatter in the data from 35 to 110° due to
the influence of the buildings on Meteorological Monitor #1. The wind speed differences
versus the Monitor #1 wind directions, shown in Figure 19, are almost all less than 0.4. The
largest wind speed differences occur around 180 and 360°.
Site #6 near the Battery Tunnel is divided into two separate sites, 6A and 6B. There is
one meteorological tower at each site, which are almost two blocks apart. As shown in
Figures 20 and 21, the wind field at this site does not appear to be uniform.
32
-------
13
£
in
s
1.5
1.0
0.3
!»-* '•
£ 0.0
5
-a.s
-1.0
3 30 180 270 360
WIND DIRECTION - MONITOR 1 (OEG3
Figure 8. Cos(0), where 0 is the wind direction difference between Monitors 1 and 2,
as a function of the wind direction at Monitor 1 for all 1-hour average data
at Site #1. .
i
13
>
-------
-l.S
90 180 270
.HIND DIRECTION - MONITOR 1 COEGJ
Figure 10. Cos(0),' where 9 is the wind direction difference between Monitors 1 and 2,
as a function of the wind direction at Monitor 1 for all 1-hour average data
at Site #2.
2.0
l.S
i.a
0.5 ;../>-;;:•' • ; .
360
0.0
0 Sa 180 270
WIND DIRECTION - MONITOR 1 (OEG)
Figure 11. The ratio of the wind speed difference between Monitors 1 and 2 to the
average wind speed at the same monitors as a function of the wind direction
at Monitor 1 for all 1-hour average data at Site #2.
34
-------
1.0
0.5
3S-*V.£.:i
•'..r1 • /t
• r*»-»*. » • -CtXj;
«•, * • «• w ^..^i
. <'> ' "-
•>v/
350
90 180 270
MIND DIRECTION - MONITOR 3 (DEG3
Figure 12. Cos(0), where 6 is the wind direction difference between Monitors 2 and 3,
as a function of the wind direction at Monitor 3 for all 1-hour average data
at Site #2.
2.2
US
C3
I
cfT
tn
1.2 -
90- 180. 270. 360.
WIND DIRECTION - MONITOR 1 COEGJ
Figure 13. The ratio of the wind speed difference between Monitors 2 and 3 to the
average wind speed at the same monitors as a function of the wind direction
at Monitor 1 for all 1-hour average data at Site #2.
35
-------
1.5
1.0
-. a.s
i
0.0
-------
1.5
1.3
3.5
£ 3.3
I '
tn
° -a. s
-1.3 -
-i.S
• -"" ..r*
.. f •
30 180 270
WIND DIRECTION - MONITOR 1 COEC3
363
Figure 16. Cos(0), where 9 is the wind direction difference between Monitors 1 and 2,
as a function of the wind direction at Monitor 1 for all 1-hour average data
at Site #4.
2
i
2.3
I.S
1.3
3.S
3.3
SB 1S0 270
HIND DIRECTION - MONITOR 1 (OEGJ
350
Figure 17. The ratio of the wind speed difference between Monitors 1 and 2 to the
average wind speed at the same monitors as a function of the wind direction
at Monitor 1 for all 1-hour average data at Site #4.
37
-------
l.S
1.0
3 0.5
a
a. a
-8.S
-l.S
_L
_L
_L
3 30 180 270 360
HINO DIRECTION - MONITOR 1 (OEGJ
Figure 18. Cos(0), where 9 is the wind direction difference between Monitors 1 and 2,
as a function of the wind direction at Monitor 1 for all 1-hour average data
at Site #5.
2.0
l.S
1.0
0.S <-
^.5^^^ .-:•- '•.'*+'"%£?*
3 30 180 270 380
HIND DIRECTION - MONITOR 1 CQEGJ
Figure 19. The ratio of the wind speed difference between Monitors 1 and 2 to the
average wind speed at the same monitors as a function of the wind direction
at Monitor 1 for all 1-hour average data at Site #5.
38
-------
-1.5
90 180 270 360
MIND DIRECTION - MONITOR 1 (OEGJ
Figure 20. Cos(6), where 9 is the wind direction difference between Monitors 1 and 2,
as a function of the wind direction at Monitor 1 for all 1-hour average data
at Site #6.
-------
The results of the uniform wind analysis (DiCristofaro et al., 1991) indicate that Sites #5
and #1 are best in terms of unhindered approach flows and wind field uniformity. The
presence of complex flows including street canyon rotors and strongly channeled flows are
indicated at the remaining sites, although the uniformity of wind speeds is good at Site #2.
3.3.2 Characterization of the Wind Speed and Stability Class for Modeled Hours
Tables 5 and 6 present tabulations of stability and wind speed, respectively for all
modeled hours. As shown in Table 5, there are more hours classified as unstable at all sites
except Sites #3, 4 and 6. If the neutral and stable hours are combined then there are more
neutral/stable hours for all sites except Sites #2 and 5. As shown in Table 6, Sites #3, 4, and
6 have predominantly light wind speeds (£ 6 mph) for almost all modeled hours.
3.3.3 Traffic Counts
The total number of vehicles modeled at each site (e.g., all modeled links at each
intersection) as a function of the model hour is shown graphically in Figure 22. On average,
there is very little variation in traffic counts from one hour to the next because most of the
traffic data are associated with rush-hour conditions. This is not surprising, since the hours
were selected on the basis of the maximum observed CO concentrations. The traffic counts
are lowest on average at Site #4 and highest at Site #6.
40
-------
TABLE 5
TABULATION OF STABILITY CLASSIFICATION BY SITE
Site
1
2
3
4
5
6
#of
Hours
142
143
66
74
75
75
Stability Classification
Unstable
62
86
24
23
39
21
Neutral .
57
44
16
34
26
36
Stable
23
13
26
17
10
18
TABLE 6
TABULATION OF WIND SPEED BY SITE
Site
1
2
3
4
5
6
#of
Hours
142
143
66
74
75
75
Wind Speed
^ 6 mph
98
121
66
73
56
75
>6
44
22
0
1
19
0
mph
41
-------
Site #1
11 I 21 I 31 I 41 | rt | 81 I 71 I 81 I 3"1 1101 1111 1121 1131 I H
I 1! 28 38 48 58 35 58 38 93 108 118 118 138
WXMIM Hour
Site #2
Site #3
'I
15
Figure 22. The number of vehicles modeled versus the consecutive model hour at each
site.
42
-------
Site *
I*
^ §
il
is
Jl
IS
1 8 11 10 21 28 31 35 41 18 51 58 51 38 71
Site #5
Site #S
1 5 11 18 21 2S 31 36 11 46 51 38 51
Figure 22. The number of vehicles modeled versus the consecutive model hour at each
site (Continued).
43
-------
-------
4.0 MODELING METHODOLOGY
4.1 Model Input Data
4.1.1 Intersection Configurations
A summary of the intersection data used to specify all modeled links at each site is
presented in Table 7. The link ID, intersection street names, the number of lanes, link width,
total modeled link length, and whether the link is an approach or departure roadway are
summarized. A link is considered to be any lane group that is considered to be a separate
line source and can be characterized separately from other sources. The link length shown in
Table 7 is with respect to the center of the intersection to the center of the adjacent
intersection.
Each of the eight models evaluated requires a slightly different characterization of the
intersection data. Overlapping free-flow and queue and/or excess emission links are required
by the EPAINT, CAL3QHC, and VOL9MOB4 modeling techniques. For these three models,
each free-flow link is modeled using the distance from the center of the adjacent intersection
to the center of the modeled intersection. The modeled queues and/or excess emission links
are modeled from the stop lines. The GIM model requires the designation of separate free-
flow and excess emission links. Each link modeled by the GIM model consists of an arrival
link at cruise emissions, an excess emission link adjoining the arrival link to the intersection
center, and a departure link in the opposite direction. If the excess emission link is estimated
by the GIM model to be greater than the overall length of the link, then an arrival portion of
the link is not modeled. The FHWAINT model only requires the midblock-to-midblock
distances over which the adjusted emissions for the vehicle speed are applied. No excess
emission links are specified in the FHWAINT model. The TEXIN2 and CALINE4 models
only require the specification of general link coordinates for each leg of the intersection. For
example, when applying the TEXIN2 model at Site #1, the West Street and Service Road
approach lanes were combined into one group of lanes. The TEXIN2, CALJNE4, and IMM
models internally calculate the link and departure components, including the excess emission
links. The IMM model requires the link approach and departure coordinates. The queue
links are internally calculated by the IMM model and are superimposed over the approach
links.
The intersection configurations for each site used in the modeling analysis are shown in
Figures 23 through 28. Only a portion of the total modeled links is shown in these figures.
The actual modeled link lengths are given in Table 7. Also shown in these figures are the
locations of the CO monitors (labeled as PI, P2, etc.) and the meteorological monitors
(labeled as Ml, M2, etc.).
The intersection configuration for Site #1 (West/Chambers) is shown in Figure 23. Nine
separate links (five approach and four departure) were used in the modeling analysis at Site
45
-------
TABLE?
SUMMARY OF INTERSECTION CONFIGURATIONS
SITE#1
Link
ID Intersection
WN-340 West NB @ Chambers
WN-410 West NB @ Main Line
WS-510 West SB @ Chambers
WS-620 West SB @ Barclay
WN-350 Service NB @ Chambers
WN-420 Service NB @ 1600'
WS-520 West SB @ Chambers (left)
CW-210 Chambers WE @ West
CW-110 Chambers EB @ Greenwich
Link
ID Intersection
TE-108 34th EB @ 8th
TE-107 34th EB @ 7th
TW-208 34th WB @ 8th
TW-209 34th WB @ 9th
EN-334 8th NB @ 34th
-EN-335 8th NB @ 35th -
Link
ID Intersection
SB-520 West 65th EB @ Broadway
SB-530 West 65th EB @ Central Park W.
SB-650 Broadway NB @ 65th
SB-660 Broadway NB @ 66th
SB-865 Broadway SB @ 65th
SB-864 Broadway SB @ 64th
SB-965 Columbus SB @ 65th
SB-964 Columbus SB @ 64th
Number
of Lanes
3
3
3
3
3
3
1
2
1
SITE
Number
of Lanes
2
2
2
2
4
4
SHE
Number
of Lanes
2
2
3
3
3
3
3
4
Width
(ft)
36
36
36
36
36
36
12
23
16
#2
Width
(ft)
24
24
24
24
48
48
#3
Width
(ft)
24
24
36
36
36
36
36
48
Length
(ft)
790
1572
1633
1023
772
1576
1643
685
. 606
Length
(ft)
920
852
914
855
300
262
Length
(ft)
882
930
355
248
345
267
258
242
Approach/
Departure
A
D
A
D
A
D
A
A
D
Approach/
Departure
A
D
A
D
A
D
Approach/
Departure
A
D
A
D
A
D
A
D
46
-------
TABLE 7 (continued)
SUMMARY OF INTERSECTION CONHGURATIONS
SITE #4
Link
ID Intersection
SF-870 57th EB @ 7th
SF-860 57th EB @ Ave of Amer.
SF-670 57th WB @ 7th
SF-675 57th WB @ Broadway
SF-570 7th SB @ 57th
SF-560 7th SB @ 56th
Link
ID Intersection
TN-340 12th NB @ W 34th
TN-390 12th NB @ W 39th
SN-340 Service NB @ W 34th
SN-390 Service NB @ W 39th
TS-340 12th SB @ W 34th
TS-300 12th SB @ W 30th
TL-340 12th SB (left turn) @ W 34th
SS-340 Service SB @ W 34th
SS-300 Service SB @ W 30th
TW-115 34th EB@ 10th
TE-110 34th SB @ 12th
Link
ID Intersection
WT-110 West NB @ Tunnel
WT-120 West NB @ Liberty
WT-125 West NB @ Liberty (left)
WT-330 West SB @ Tunnel Underpass
WT-320 West SB @ Tunnel
WT-310 West SB @ Morris
WT-410 Service SB <® Morris
WT-500 Tunnel WB @ West
WT-510 Tunnel EB (Entrance)
Number
of Lanes
2
2
2
2
4
4
SITE
Number
of Lanes
2
2
3
2
2
2
1
1
1
2
3
SITE
Number
of Lanes
4
3
1
2
3
2
2
5
2
Width
(ft)
24
24
24
24
48
48
#5
Width
(ft)
24
24
30
24
24
24
12
12
-12
24
36
#6
Width
(ft)
48
36
11
24
36
24
24
60
25
Length
(ft)
631
844
939
525
315
252
Length
(ft)
1174
1174
1177
1177
1179
1179
1342
1172
1172
995
912
Length
(ft)
257
1330
1318
1104
1487
221
221
509
474
Approach/
Departure
A
D
A
D
A
D
Approach/
Departure
A
D
A
D
A •
D
A
A
D
A
D
Approach/
Departure
A
D
D
A
A
D
D
A
D
47
-------
o
I
v«^
T—i
=*
CO
I
•*-*
I
o
.2
•j~»
o
-------
60
1
en
•a
.1
1
«50
'43
O
o
_o
*4n»
O
I
« ca
P §
s
fi
49
-------
•s
»v=0
e/3
3
->;
2
§
60
<£!
5
u
o
£
<3 c«
C °w
•" >.
2 "3
H I
CNI
2
s
aa
50
-------
.2
•o
s
«n
§
•a
S
o
*•*
_o
o
S ca
-------
Bffl
I
.s
•o
CO
3
I
1
to
iS
§
o
.2
1
e
s
&a
£
52
-------
-f-
P1
N
0 10 20 30
Figure 28. The intersection configuration for Site #6 (Battery Tunnel) used in the modeling
53
-------
#1. Before approaching the West/Chambers intersection, the northbound portion of West
Street divides in two with the formation of a Service Road. The southbound left-turn lane on
West Street was modeled as a separate link. The configuration shown for Site #2 (34th/8th)
in Figure 24 consists of four one-way northbound lanes on 8th Avenue and two approach and
two departure lanes on West 34th Street. The Site #3 (65th/Broadway) configuration shown
in Figure 25 is the most complicated intersection of the six modeled. The intersection
consists of four approach links (Columbus Avenue southbound, Broadway north and
southbound, and West 65th Street eastbound) and four departure links. The intersection
configuration for Site #4 (57th/7th), shown in Figure 26, consists of four one-way southbound
lanes on 7th Avenue and two approach and two departure lanes on West 57th Street. The
Site #5 (34th/12th) configuration shown in Figure 27 indicates that there are six different
approach links and five departure links used in the modeling analysis. On this portion of
Route 9A, 12th Avenue includes a separate north and southbound service road which were
modeled as separate links. The southbound left-turn lane on 12th Avenue, was modeled as a
separate link. Finally, the intersection configuration for Site #6 (Battery Tunnel), shown in
Figure 28, consists of four approach links and five departure links. The WT-330 link
represents the Route 9A entrance to the Brooklyn Battery Tunnel and the WT-500 link
represents the tunnel exit. The traffic associated with the WT-500 link must pass under an
overhang associated with the Port of New York Authority Building before intersecting West
Street
4.1.2 Traffic and Emissions Characterization
As discussed in Section 3.0, the approach and departure traffic counts for each modeled
link were obtained by manual processing of the videotapes. The capacity or saturated flow to
each approach of the intersection was calculated using the computerized version of Chapter 9
of the 1985 Highway Capacity Manual (TRB, 1985). The actual average green time for each
signal phase was used along with an average (or random) arrival progression factor. Average
traffic volumes and turn information (left, thru, right) for each link were used along with the
average percent red times. The percent red time for four periods during the day along with
the calculated average saturated flow rates for each link are shown in Table 8. The percent
red time and cycle time were used to calculate the green time for four different time periods
for those models requiring the input of green time. Yellow time was assumed to be zero for
aU analyses. The segment running time per mUe was input to the EPAINT model using the
recommended values in HCM (TRB, 1985) based on the cruise speed and the average
segment length.
Overcapacity conditions were modeled by the EPAINT, FHWADMT, and VOL9MOB4
models at a few links and sites. Overcapacity conditions exist when the fraction of vehicles
that stop is greater than one in the VOL9MOB4 model and when the volume to capacity
(V/Q ratio is greater than 1.2 in the EPAINT and FHWAESfT models. When overcapacity
conditions were modeled at a particular link, the respective model did not produce any
emissions for that link. Thus, in order to compensate for overcapacity conditions, the
54
-------
TABLES
SUMMARY OF SATURATED FLOW AND PERCENT RED TIME
SITE#1
Link
ID
WN-340
WN-410
WS-510
WS-620
WN-350
WN-420
WS-520
CW-210
CE-110
Link
ID
TE-108
TE-107
TW-208
TW-209
EN-334
EN-335 '
Link
ID
SB-520
SB-530
SB- 650
SB-660
SB-865
SB-864
SB-965
SB-964
Saturated
Flow
(vehicles)
4811
-999
4811
-999
4768
-999
1524
3019
-999
SITE
Saturated
How
(vehicles)
2149
-999
2688
-999
6205
-999
SITE
Saturated
Flow
(vehicles)
2688
-999
4321
-999
4099
-999
4199
-999
Percent Red Time
Time Period (EST)
5-9 10-14 15-18
0.533 0.408 0.642
0.275 0.275 0.275
0.266 0.275 0.417
0.325 0.325 0.325
0.533 0.408 0.642
0.275 0.275 0.275
0.733 0.867 0.775
0.766 0.766 0.625
0.529 0.529 0.529
#2
Percent Red Time
Time Period (EST)
5-9 10-14 15-18
0.572 0.572 0.572
0.569 0.569 0.569
0.572 0.572 0.572
0.569 0.569 -0.569
0.470 0.470 0.470
0.412 0.412 0.412
#3
Percent Red Time
Tune Period (EST)
5-9 10-14 15-18
0.670 0.670 0.670
0.679 0.679 0.679
0.670 0.670 0.670
0.419 0.419 0.419
0.670 0.670 0.670
0.330 0.330 0.330
0.660 0.660 0.660
0.346 0.346 0.346
19-4
0.408
0.275
0.275
0.325
0.408
0.275
0.867
0.766
0.529
19-4
0.572
0.569
0.572
0.569
0.470
0.412
19-4
0.670
0.679
0.670
0.419
0.670
0.330
0.660
0.346
Note: Saturated flow values of -999 indicate departure links.
55
-------
TABLE 8 (continued)
SUMMARY OF SATURATED FLOW AND PERCENT RED TIME
SITE #4
Link
ID
SF-870
SF-860
SF-670
SF-675
SF-570
SF-560
Link
ID
TN-340
TN-390
SN-340
SN-390
TS-340
TS-300
TL-340
SS-340
SS-300
TW-115 '
TE-110
Link
ID
WT-110
WT-120
- WT-125
WT-330
WT-320
WT-310
WT-410
WT-500
WT-510
Saturated
Flow
(vehicles)
2755
-999
2637
-999
6324
-999
SITE
Saturated
Flow
(vehicles)
3208
-999
4234
-999
3208
-999
1524
1604
-999
2430
-999
SITE
Saturated
Flow
(vehicles)
6144
-999
-999
3600
4811
-999
-999
6961
-999
Percent Red Time
Time Period (EST)
5-9 10-14 15-18
0.608 0.608 0.608
0.632 0.632 0.632
0.606 0.606 0.606
0.571 0.571 0.571
0.461 0.461 0.461
0.411 0.411 0.411
#5
Percent Red Time
Time Period (EST)
5-9 10-14 15-18
0.525 0.52S 0.525
0.193 0.375 0.375
0.525 0.525 0.525
0.193 0.375 0.375
0.208 0.375 0.375
0.250 0.250 0.250
0.717 0.858 0.858
0.208 0.375 0.375
0.250- 0.250 0.250
0.833 0.667 0.667
0,733 0.733 0.733
#6
Percent Red Time
Time Period (EST)
5-9 10-14 15-18
0.567 0.422 0.422
0.508 0.508 0.508
0.508 0.508 0.508
0.000 0.000 0.000
0.567 0.422 0.422
0.433 0.433 0.433
0.000 0.000 0.000
0.467 0.622 0.622
0.000 0.000 0.000
19-4
0.608
0.632
0.606
0.571
0.461
0.411
19-4
0.525
0.375
0.525
0.375
0.375
0.250
0.858
0.375
0.250
0.667
0.733
19-4
0.422
0.508
0.508
0.000
0.422
0.433
0.000
0.622
0.000
Note: Saturated flow values of -999 indicate departure links.
56
-------
overcapacity link was merged with another link or the volume was set to capacity. For
example, at Site #5, the left turn lane (TL-340) was modeled as overcapacity by EPAINT for
five different hours. When this link (TL-340) was merged with TS-340, the V/C ratio was
less than 1.2 and emissions from the link were calculated.
When applying the CALINE4 model, the amount of time the first car spends in the
queue (IDT1) was set to the red time and the vehicle idle time at the end of the queue (IDT2)
was set to zero (Benson, 1991). The value of NDLA, the length of the queue or the number
of cars per lane that are queued when the light turns green, input to the CALINE4 model was
calculated using the following steps:
(1) NDLA = (number of vehicles/number of lanes) • percent red time
(2) "Ripple" or Propagation Tune of the Queue = NDLA • 2.5 sec/car
(3) Adjusted Percent Red Time = (red time + propagation time)/cycle time
(4) Adjusted NDLA = (number of vehicles/number of lanes) • adjusted percent red time
The value of NDLA was adjusted in order to account for a "ripple" or propagation speed
estimated at 2.5 sec/vehicle. Also, in the CALINE4 model, if the calculated length of the
queue plus the deceleration length is greater than the link length with respect to the stop line,
then the model will stop with an error. Therefore, for these traffic conditions, the link length
was reset to the length of the queue plus the deceleration length.
The hourly traffic data contain four types of vehicle speeds for each link: cruise speed
on the block, cruise speed on the downstream block, average speed, and "modified" average
speed. The "modified" average speed is the total travel time less the average stop delay time
on the link. For this modeling analysis, the cruise speed on the block for each link was used
for the approach speed. For the TEXIN2 model, a traffic volume weighted average of the
cruise speeds over the modelled lane group was used. For those models which require a
departure speed, we specified the cruise speed on the downstream block associated with the
approach lane being modeled. The traffic cruise speeds for each link modeled are presented
in Table 9 for four different time periods. At Site #2 there were a few exceptions to the
traffic speeds used with respect to the time period.
One of the purposes of this model evaluation study is to evaluate the intersection
modeling techniques using commonly available data. Therefore, the cruise speed rather than
the average or "modified" average speed was used in the evaluation. Also, the observed
queue lengths were not used so that the queuing algorithms for each model could be tested.
Emissions were estimated using the MOBILE4 and MOBILE4.1 emissions model. The
Inspection/Maintenance (I/M) program specifications for the MOBELE4 and 4.1 modeling
were set as follows:
57
-------
TABLE 9
TRAFFIC CRUISE SPEEDS (MPH) USED IN MODELING ANALYSIS
SITE#1
TJnlf
WN-340
WN-410
WS-510
WS-620
WN-350
WN-420
WS-520
CW-210
CE-no
Time Period (EST)
6-9
293
30.4
342
30.5
293
30.4
34.2
18.5
20.4
10-14
28.9
32.8
31.4
28.8
28.9
32.8
3L4
17.2
20.1
15-18
3L8
27.0
31.9
2L8
31.8
27.0
31.9
173
16.7
19-5
36.1
40.8
39.4
38.4
33.9
35.7
373
22.0
24.1
SITE #2
Link
TE-108
TE-107
TW-208
TW-209
EN-334
EN-335
Exceptions:
7-9
112.
20.1
20.8
20.6
ZL5
222.
10-25-89 Hr.
U-08-89 Hr.
Time Period
10-14
12.9
7.6
18.5
23.2
9.8
10.8
18 9-16-89 Hr. 19
15
(EST)
15-18
123
12.4
16.0
20.9
133
15.5
8-17-89 Hr. 8
19-6
25.6
30.6
30.5
313
33.7
27.9
8-14-89 Hr. 8
10-27-89 Hr. 7
SITE #3
Link
SB-520
SB-530
SB-650
SB-660
SB-865
SB-864
SB-965
SB-964
7-9
213
18.1
17.5
233
18.7
21.9
17.6
18.2
Time Period
10-14
193
15.9
14.2
17.8
16.2
18.0
15.1
20.0
(EST)
15-18
213
16.0
15.9
20.4
17.2
193
16.8
183
19-6
30.1
20.8
13.2
18.2
16.2
26.2
29.5
29.0
58
-------
TABLE 9 (continued)
TRAFFIC CRUISE SPEEDS (MPH) USED IN MODELING ANALYSIS
SITE #4
Link
SF-870
SF-860
SF-670
SF-675
SF-570
SF-560
7-9
24.0
23.5
23.6
18.7
12.6
18.5
Time Period
10-14
20.5
243
17.0
143
18.8
19,9
(EST)
15-18
20.8
17.6
22.8
23.2
15.9
173
19-6
32.8
242
25.1
23.6
20.1
232
SITE #5
Link
TN-340
TN-390
SN-340
SN-390
TS-340
TS-300
TL-340
SS-340 •
SS-300
TW-115
TE-110
7-9
31.0
29.5
32.0
28.5
32.8
32.9
32.8
35.1
37.8
25.8
25.4
Time Period
10-14
34.7
33.6
31.4
30.5
28.6
3LO
28.6
33.5
375
27.7
25.0
(EST)
15-18
28.4
22.9
28.4
22.9
27.8
27.5
27.8
30.9
34.1
27.5
23.7
19-6
38.1
38.5
353
34.4
36.0
393
36.0
42.0
43.5
29.8
27.6
SITE #6
Link
WT-110
WT-120
WT-125
WT-330
WT-320
WT-310
WT-410
WT-500
WT-510
7-9
15.9
193
19.3
30.6
30.6
29.0
29.0
343
36.6
Time Period
10-14
15.2
283
283
30.9
30.9
30.4
30.4
44.5
43.5
(EST) •
15-18
15.0
29.1
29.1
33-5
33.5
283
283
34.2
30.7
19-6
29.1
35.8
35.8
35.1
35.1
34.4
34.4
46.1
47.8
59
-------
• Start year - 1982
• Pre-1981 MYR stringency rate - 30%
• First model year covered - 1960
• Last model year covered - 2020
• Waiver rate (pre-1981) - 0.0%
• Waiver rate (1981 and newer) - 0.0%
• Compliance rate - 75%
• Inspection type - Manual decentralized
• Inspection frequency - Annual
• Vehicle types covered - LDGV, LDGT1, LDGT2, HDGV
• 1981 and later MYR test type - Idle
The Anti-Tampering Program (ATP) program specifications for the MOBHJE4 and 4.1
modeling were:
• Start year - 1984.
• First model year covered - 1960
• Last model year covered - 2020
• Vehicle types covered - LDGV, LDGT1, LDGT2, HDGV
• Type - Decentralized
• Frequency - Annual
• Compliance Rate - 75%
• Air pump system disablements - Yes
• Catalyst removals - Yes
60
-------
• Fuel inlet restrictor disablements - No
• Tailpipe lead deposit test - No
• EGR disablement - Yes
• Evaporative system disablements - No (Yes for MOBELE4.0)
• PCV system disablements - Yes
• Missing gas caps - No
The MOBILE4.1 model will only model an ATP with an evaporative system inspection and
provide appropriate emission credits if a gas cap inspection is also included. If the user
indicates that an evaporative system inspection is performed, but that a gas cap inspection is
not performed, an error message will be issued and execution of the run will stop (EPA,
1991). The New York DEP (Nudelman, 1991) recommends not using the evaporative control
systems check when using MOBILE4.1.
Mileage accumulation rates and registration distributions.recommended by the New York
DEC for automobiles are listed in Table 10. The MOBILE4.1 emissions model requires an
additional five years of data (Years 21 to 25) for the mileage accumulation rates and
registration distributions. Data for Years 21 to 25 were not available from the New York
DEC when the MOBILE4.1 modeling was conducted. As recommended by the New York
DEP (Nudelman, 1991), the values for years 21 to 25 were set to zero.
As noted hi Section 3.0, hourly vehicle mixes were available for each link for seven
vehicle categories: automobiles, fleet medallion New York City taxis, non-fleet medallion
New York City taxis, non-medallion New York City taxis, light-duty trucks, heavy-duty gas
trucks, and heavy-duty diesel trucks. The mileage accumulation rates, registration distri-
butions, I/M. (Inspection/Maintenance) program parameters, and ATP (Anti-Tampering Pro-
gram) parameters are different for each vehicle category. A large percentage of the vehicles
in Manhattan are taxis which differ from automobiles in the following manner:
1) The taxi turnover rate is high so the vehicles are newer. Thus, taxis tend to have
installed more current control technologies.
2) Taxis are constantly cruising so they are almost always hot; whereas, the thermal
states of automobiles vary during the day.
3) Taxis have been subject to the I/M program longer than automobiles. Also, the I/M
program is more strict for taxis than automobiles.
4) In late 1989, taxis were subjected to centralized inspections three times per year.
61
-------
s
I
m
cs
S
O
SO
i
oi
_o
J2
o
8?
-------
CO
55
I
i
H
CO
331
|
1
9
a
S£
m
o
o
1—1
s
ffl
<•
P
5
co •"»
11
^s
88
^
IQ
Is
§
•a
1
CO
5
O
"1
CO
'5a
-------
The idle emission factors from taxis are significantly lower than those from other
automobiles. For conservative modeling purposes, no adjustments were made to account for
lower emissions from taxis. Thus, for MOBDLE4 and MOBILE4.1 modeling purposes, the
auto mileage accumulation rates, registration distributions, I/M program parameters, ATP
parameters, and refueling loss parameters for automobiles were used. The hourly vehicle
mixes for each link were combined in the following manner:
LDGV (Light Duty Gas Vehicles) = automobiles + fleet medallion taxis + non-fleet
medallion taxis + non-medallion taxis
LDDT (Light Duty Diesel Trucks) = 1.8% Light Duty Trucks for MOBILE4
0.8% Light Duty Trucks for MOBILE4.1
LDGT1 (Light Duty Gas Trucks Category 1) = 58.7% Light Duty Trucks for MOBILE4
67.8% Light Duty Trucks for MOBILE4.1
LDGT2 (Light Duty Gas Trucks Category 2) = 39.5% Light Duty Trucks for MOBILE4
31.4% Light Duty Trucks for MOBILE4.1
HDGV (Heavy Duty Gas Vehicles) = heavy duty gas trucks
MC (Motorcycles) = 0
LDDV (Light Duty Diesel Vehicles) = 0
The factors used to calculate the LDDT, LDGT1, and LDGT2 distributions are based on the
MOBILE4 and MOBILE4.1 default values for 1989.
The thermal state conditions for each modeled link were obtained from field interviews
during the monitoring program for four different time periods. The percent cold thermal
states used in the modeling analysis are presented in Table 11. For most models, the thermal
state conditions were input as a function of the link. However, for the IMM model, a traffic
volume weighted average was used to calculate single values for the percentage of hot starts
and percentage of cold starts for the hour being modeled. Similarly, the TEXIN2 model only
allows the input of thermal state conditions based on the lane segment groups input to the
model. The modeling analysis assumes that there were no hot starts and the catalytic and
non-catalytic cold starts were equal.
year:
The Reid Vapor Pressure (RVP) and ASTM class were specified based on the time of
2/89 - 5/89 RVP = 11.5 psi (MOBILE4); ASTM = C
RVP = 11.9 psi (MOBILE4.1)
64
-------
TABLE 11
THERMAL STATES (% COLD) USED IN MODELING ANALYSIS
SITE#1
Link
WN-340
WN-410
WS-510
WS-620
WN-350
WN-420
WS-520
CW-210
CE-110
5-9
6.9
6.9
63
6.9
6.9
6.9
6.9
6.9
6.9
Hour
10-14
12.1
12.1
12.1
12.1
12.1
12.1
12.1
12.1
12.1
Range (EST)
15-18
17.0
17.0
3.4
3.4
17.0
17.0
12.7
7.8
12.7
19-4
18.8
18.8
4.5
4.5
18.8
18.8
13.1
13.1
13.1
SITE #2
Link
TE-108
TE-107
TW-208
TW-209
EN-334
EN-335
5-9
6.9
6.9
6.9
6.9
8.5
8.5
Hour
10-14
5.4
8.1
14.0
14.0
12.1
12.1
Range (EST)
15-18
8.0
8.0
15.0
15.0
12.1
12.1
19-4
15.1
15.1
15.1
15.1
19.0
19.0
SITE #3
Link
SB-520
SB-530
SB-650
SB-660
SB-865
SB-864
SB-965
SB-964
5-9
10.0
153
16.9
16.9
14.0
183
25.0
21.6
Hour
10-14
6.0
12.6
20.8
20.8
16.1
16.1
20.0
173
Range (EST)
15-18
23.4
23.4
25.0
25.0
21.2
21.2
24.4
24.4
19-4
23.1
23.1
18.1
18.1
22.9
22.9
223
223
65
-------
TABLE 11 (continued)
THERMAL STATES (% COLD) USED IN MODELING ANALYSIS
SITE #4
Link
SF-870
SF-860
SF-670
SF-675
SF-570
SF-560
5-9
143
143
13.0
13.0
113
113
Hour
10-14
9.4
9.4
53
53
13.5
13.5
Range (EST)
15-18
253
253
95
95
183
183
19-4
25.1
25.1
8.9
8.9
15.6
15.6
SITE #5
Link
TN-340
SN-340
TS-340
TL-340
SS-340
TW-11S
TE-110
5-9
13.6
13.6
7.5
7.5
15
22.0
8.0
Hour
10-14
18.9
18.9
8.5
8.5
8.5
18.0
6.0
Range (EST)
15-18
14.1
14.1
15.4
15.4
15.4.
26.0-
14.0
19-4
9.8
9.8
6.7
6.7
6.7
19.0
1LO
SITE #6
TJnlf
WT-110
WT-120
WT-125
WT-330
WT-320
WT-310
WT-410
WT-500
WT-510
Hour Range (EST)
5-9
7.0
3.8
3.8
15.0
15.0
8.2
8.2
0.0
11.6
10-14
15.0
9.6
9.6
15.0
15.0
10.0
10.0
0.0
15.0
15-18
27.0
17.5
17.5
17.0
17.0
13.6
13.6
0.0
22^
19-4
12.0
7.7
7.7
18.0
18.0
12.0
12.0
0.0
15.5
66
-------
8/89 - 9/89 RVP = 9.0 psi; ASTM = A
After 9/89 RVP = 11.5 psi; ASTM = C
RVP = 11.9 psi (MOBILE4.1)
The RVP values are different for MOBHJE4 and MOBILE4.1 because the accepted maximum
input values are different for the two different versions of the emissions model.
MOBILE4 idle emissions are calculated at 75°F, 0% cold starts, 0% hot starts, 2.5 mph,
and 9.0 psi RVP. In order to adjust the MOBILE4 idle emissions to the scenario conditions
being modeled, and idle correction factor (ICF) was calculated by dividing the composite
MOBTLF.4 emission factor at the condition of interest by the composite MOBILE4 emission
factor based on the MOBILE4 idle condition, assuming a travel speed of 2.5 mph:
MOBILE4Scenono(yearyhotlcold%,temp,RVP,2.5mpK)
MOBlLE4Idle (year,0/0/0,75F,5KP=9.0,2.5 mph)
(1)
The idle emission factor in g/veh min for the models utilizing an idle correction factor
(CAL3QHC, EPAINT, CALINE4, VOL9MOB4, and IMM) was calculated using:
Idle Emission =
MOBILE4Idle X ICF
60
(2)
The idle adjustment is performed automatically by the MOBHJE4.1 emissions model, so
no external corrections were needed for MOBHJ24.1. It should be noted that there is a major
difference between the MOBILE4 and MOBILE4.1 versions of TEXIN2 because the
MQBTLF.4 version did not include an idle correction and the most recent version
(MOBILE4.1) automatically includes the idle correction.
4.1.3 Meteorological and Background Data
The hourly-averaged temperature data from the meteorological towers at each site were
averaged to calculate a site-specific hourly value. For the remaining meteorological input
data (wind speed, wind direction, sigma theta, stability class), the meteorological tower closest
to the CO monitor location was used. Mixing heights of 1000 m were used, since the results
are not affected if the mixing height is between 100 and 1000 m high and mixing heights
below 100 m in Manhattan rarely occur.
In addition to the use of the observed meteorological data, a subset of hours (top 10 at
each site) were modeled using the regulatory default meteorological conditions described
below:
67
-------
Wind Speed = 1 m/s
Stability Class = D
Sigma-Theta (ae) = 25°
Observed Temperature
"Worst-Case" Wind Direction Angle (determined using 10° increments)
The closest background concentration (Battery Park or the Post Office Station) was
subtracted out of the observed concentration at the monitors at each intersection site. All
modeling was performed for one hour averages only. After the removal of the background
concentrations, a screening threshold of 0.5 .ppm was used. For example, when both the
observed and predicted concentrations at a monitor are less than 0.5 ppm, that data pair was
eliminated from the data set
4.2 Dispersion Modeling Techniques
Atmospheric dispersion of the vehicular CO emission at each site was simulated using
either the CALINE3 or HIWAY2 models. Only IMM4 uses the HIWAY2 dispersion
methodology; all the other models evaluated use the CALINE3 model. As discussed in
Section 2.0, the CALINE3 or HIWAY2 dispersion algorithms are included in the CAL3QHC,
TEXIN2, CAUNE4, and IMM models. The CALINE3 model was run separately for the
EPAINT, FHWAINT, VOL9MOB4, and GIM modeling techniques.
Each modeled roadway was divided into free-flow traffic links and queuing or excess-
emission links, as required for each model. As recommended in the CALINE3 User's
Manual, an additional six meters (three meters for each side of the roadway) was added to the
actual width for CALINE3 dispersion modeling to account for wake-induced plume
dispersion. Turbulence is assumed to be zero for queued vehicles, so no additional width was
added for the queuing links. All mobile source heights were modeled at 0.0 m and all links
were assumed to be at grade.
A surface roughness length of 3.21 m was used for approach flows over numerous city
blocks. Lower values of the surface roughness length (0.03 m) were used at Site #5
(34th/12th) when the intersection was exposed to flows over the Hudson River without
intervening buildings. For modeling CO concentrations, the settling velocity and deposition
velocity were set at zero because CO is a gaseous emission. An averaging time of 60
minutes was used. The CO probe heights for each CO monitor listed in Table 3 were used.
Finally, a temperature-sensitive conversion of the modeled concentrations from mg/m3 to parts
per million (ppm) was conducted.
In order to facilitate the model evaluation study and to minimize model input errors, a
series of computer programs were written for each model. The resulting "RUN" programs
read all input data from a series of standard files for each site (i.e., meteorological file, hourly
traffic data, site information file, etc.); prepare the necessary input files for all models; invoke
68
-------
the models; and list the results. For example, the "RUN" program for the GIM model, called
RUNGIM, performs the following steps for each modeled hour:
1) Set Up the MOBILE4 input file
2) Run MOBILE4
3) Set Up the GIM input file
4) Run GIM
5) Set Up the CALINE3 input file using the MOBILE4 and GIM results
6) Run CALINE3
7) Output the hourly predicted CO concentrations
69
-------
-------
5.0 STATISTICAL EVALUATION PROTOCOL
Two types of statistical evaluations of differences between observed and modeled CO
concentrations were performed. First, the EPA Model Evaluation Support System (MESS)
was used to calculate a standard set of performance measures and statistical estimators.
Second, a scoring scheme recommended by the EPA was used to rank the models and to
evaluate the significance of the results. Results of the statistical analyses are presented in
Section 6.0 and the Appendices.
5.1 The Model Evaluation Support System (MESS)
The Model Evaluation Support System (MESS) (EPA, 1987), a computerized system that
EPA uses for evaluating the accuracy of air quality models, was used to generate some of the
statistics presented in this report The Statistical Evaluation Subsystem or SES was used to
calculate the standard set of performance measures and statistical estimators as recommended
by the AMS Workshop on Statistical Data Analyses (Fox, 1981). Both paired and unpaired
data sets were used.
5.1.1 Paired Statistics
For paired comparisons, the performance measures are based on an analysis of
concentration residuals either paired in time, paired in space, or paired in both time and
space. A summary of the paired statistics which were generated by SES for each site and set
of data is given in Table 12. A select group of these statistics are presented in Section 6.0
and the Appendices. .For each site, the statistical analyses were performed for the highest
observed and predicted values for each hour (paired in time but not in space). Also, the
highest observed and predicted concentrations at each monitor for each site (paired by
monitor but not in time) were grouped for statistical analysis. Fully paired comparisons
(space and time) were made for observed and predicted concentrations at each monitor and
for each meteorological data subset Model bias is indicated by the average,
Bias = -
(3)
with a value of zero representing no bias. In Equation 3, di is the residual defined as the
observed concentration (C0) minus the predicted concentration (Cp) for the i^ data pair. The
measures of noise or scatter are computed using the following:
71
-------
r— <
W
>J
«
S
S
CO
p
<
E-
co
^
co 2
W o
ICE MEASUR
)/OR LOCATI
3 H
M ^J
S
O is
^r. ^5
2 H
PH g
|s
D P-
O
HjJ
O
S
1
"cS
O ,- Ot JS
S § co S 8
« « c .S ^
IS CO 0 .S cl
co
G £3
> JS O ""*
py^ CJ B*i *W
"-" *>'§•! S
< « S p,-a
o ^ « .2
s •§ g o
•< o,'a J2
,
•« >»
•«-j G .Q j;
**" H •-• S
S 0^.2
4-1 _ .
U G S
w P '"
43 ^* ^3
>^5 Ci< C^H *-Z3
SS S S
X XX X X XX X X
*~?- ""T* O^ -^
*™^ ^^ ^^ ^""^
y,y, y, y.
XXXXX XXX X
l""i "~i '"i *™!
XXXXX XX X X
2* ^ 2^
y. y y
XXX, X XXX X
*
l^l* ]*"•* ^ o
y y, y o
XXXXX XXX >?
CO 4^
•rt ^ *
'Q ^^ 2 ^ 'o
« S 1 1 § ®*
o 53 53 * c o e
° T3 T3 o "^ O O
i- « u c o -a cu
U
o
1
•3 J3
8 °
S <
r2 ^
o 2°
U U
„
O\ II
II Q
U" <
* :
12
-------
Variance - ^P (dt-d)2
tf-1 £1
(4)
"" " N
Gross or Mean Square Variability = — ^ j2
(5)
Average Absolute Residual = —£ \d%\
(6)
where d is the average residual (bias), and N is the number of data pairs. When the bias
approaches zero due to the cancellation of over- and underpredictions, the average absolute
residual or error can be a more meaningful measure.
For the paired comparisons, SES estimates confidence intervals on the average residual
by means of the one-sample t-test. This parametric test incorporates the assumption that the
residuals follow a normal distribution. As is discussed in Section 5.2, a bootstrap resampling
technique is also used to generate confidence intervals. The bootstrap technique yields a non-
parametric statistic because it does not invoke any assumptions regarding the statistical
distribution of the data.
5.1.2 Unpaired Statistics
For unpaired comparisons, fewer performance statistics are used. The statistical analyses
generated by SES for the N (where N = 25) highest observed and predicted values, regardless
of time or location, are summarized in Table 13. The statistics are calculated for each site
and each set of data (i.e., all hours, uniform wind hours, meteorological subsets). The
statistics for the uniform wind hours and meteorological subsets are not presented in this
report. Model bias is calculated as the difference between the average observed value and the
average predicted value. The ratio of the variances of the observed and predicted values are
calculated to indicate whether the distribution of values in the data sets are comparable. The
frequency distribution of the observed values are compared with the predicted concentrations.
5.1.3 MESS Analysis Products
As part of MESS, the Standardized SAS Graphics Subsystem or SSGS is used to provide
additional statistical tables and graphic displays of selected performance statistics. SSGS
generates output for two general classes of applications. The A-type application sorts the
input data for each group and uses only the high-25 concentrations for analysis. The A-type
73
-------
CO
3
1
CO
y
|f
^5
9 8
_
CO {13
< CO
0 O
SRFORMAN
IMPAIRED)
>-M ta-^
CU <=-'
o f-4
PI
si
8?
J 2
., o
H -a
II
p co
£3
c
o
O co
P
!> 0
°
Difference
of Averages
aH
2^
^ 2
&-
|1
— ;
y.
X
*
H-J
X
X
X
.2 g
2 o
tS
13 *c3
< "
CJ,
X
X
X
M
X
X
X
X
ce
. S
03
co g .0
| 11
CO S 8
at
I
-------
statistics are not paired in space or time, and include results for the following data groups:
• A composite of all receptors;
• Each individual receptor;
• Each stability class;
• Each wind speed class (e.g., high: u > 4 m/s; medium: 2 < u £ 4 m/s;
low: u < 2 m/s);
The statistics for each individual receptor, stability class, and wind speed class are not
presented in this report
The B-type application uses all concentrations above a selected threshold (0.5 ppm in
this evaluation) for each data group. The B-type data are paired, so a larger and more
comprehensive list of statistical comparisons are generated. All of these statistics are
calculated for each of the A-type data groups described above.
In addition to the tabular displays of selected performance statistics for each of the A
and B-type statistics, the following graphical displays were generated:
• Bias of the standard deviation versus the bias of the averages for the A (High 25) and
B (All data) type statistics.
• Quantile-quantile plots in which the high 25 observed and predicted concentrations
were plotted against each other. Since each of the 25 values is displayed for each
model, this graph is useful for detecting both the overall model bias and the points at
which the model performance is especially good or bad.
• Concentration versus cumulative frequency with observed plus multiple models per
plot Only concentrations above 50 percent frequency were plotted. On a site-by-site
basis, these plots are useful in evaluating the overall performance of each model.
5.2 Model Evaluation Scoring Scheme
5.2.1 Screening Test
The EPA (Cox, 1988) has suggested the use of a screening test for model performance,
which would normally be applied to reduce the number of models evaluated using refined
methods. This screening test was applied to the results obtained during phase I of this study,
in which MOBHJE4.0 rather than MOBILE4.1 was used to estimate emissions. The
performance measure used for the screening test is the absolute fractional bias defined as
75
-------
AFS =\FB\ = 2
(OB - PR)
(OB + PR)
(7)
where OB and PR refer to the averages of the observed and predicted highest 25 values
matched only by rank. The absolute fractional bias of the standard deviation is also used
where OB then refers to the standard deviation of the 25 highest observed values and PR
refers to the standard deviation of the 25 highest predicted values. If AFB tends to exceed
0.67 (factor of two) for either the average or the standard deviation, consideration may be
given to excluding that model from further evaluation due to its limited credibility for refined
regulatory analysis. In this evaluation of intersection models for CO, we ranked the eight
techniques by AFB in order to help indicate which of the models would be evaluated in phase
II of the study, in which MOBTLE4.1 is used to estimate emissions.
5.2.2 Refined Evaluation
The U.S. EPA has developed a method for aggregating component results of model
performance into a single performance measure that may be used to compare the overall
performance of two or more models (Cox, 1988; Cox and Tikvart, 1990). The bootstrap
resampling technique (Efron, 1982) is used to determine the significance of differences in
composite performance between models. Results from different data bases are combined
using a technique related to meta-analysis to produce an overall result.
The EPA's scoring system for refined evaluations is divided into two separate
components. The "scientific or diagnostic component" refers to the evaluation of peak
concentrations during specific meteorological conditions at each monitor and the "operational
component" refers to -the evaluation of peak averages independent of meteorological condition
or spatial location. The averages evaluated in the operational component are those for which
regulatory standards must be met (e.g., 3-hour and 24-hour averages). The capability of
models to predict concentrations at specific locations and meteorological conditions subject to
the limitations of the data base is tested using the scientific component. The New York City
data base contains mostly non-consecutive, one-hour observations, thereby limiting the
evaluation to one-hour averages. There is a regulatory standard for one-hour average CO
concentrations, so the dataset allows both diagnostic and operational components to be
evaluated. Typically, monitors are located adjacent to an intersection, so that they record
near-field concentrations during varying meteorological and traffic conditions. A diagnostic
evaluation could focus on aspects of the performance that are related to wind speed, wind
direction, stability, and traffic counts, for example. However, the wind direction aspect will
not be addressed in this evaluation. In essence, it is believed that uncertainties in the "true"
wind direction, coupled with a sparse monitoring network and a distributed source
(intersecting line sources), preclude any attempt to accurately delineate the ability of a model
to reproduce spatial relationships contained 'in the measured concentrations. Instead, the
performance will be evaluated only on the basis of the peak modeled and observed
76
-------
concentrations during each hour at each intersection. This choice eliminates any difference
between datasets for a diagnostic and an operational evaluation.
Blocking
Several subsets of the dataset for each site are formed in order to block the data
according to parameters related to significant modeling variables. Thus, differences in model
performance under different model regimes may be assessed. In this case, the relevant
parameters are wind speed, stability class, and traffic counts (a crude measure of emission
rate). The total number of vehicles modeled at each site as a function of the model hour is
presented in Figure 22 in Section 3.0. There is very little variation in traffic counts from one
hour to the next because most of the traffic data are associated with rush-hour conditions.
This is not surprising, since the hours were selected on the basis of the maximum observed
CO concentrations. Since there is not much variation in the traffic data, this parameter does
not appear to be useful when examining the scientific component The wind speeds from the
tower closest to the monitor with the maximum observed concentration are tabulated for each
site in Table 6 in Section 3.0. Sites 3, 4, and 6 have an uneven distribution of wind speeds
compared with the other three sites. The stability classifications for each site are shown in
Table 5 in Section 3.0. The stabilities seem to be more evenly distributed across each class
and over all sites.
The combined wind speed (using 6 mph or 2.7 m/s) and stability classifications are
presented in Table 14 for the three sites used in the MOBILE4.1 evaluation: Sites #1, 2, and
5. A wind speed of 6 mph was chosen in order to ensure a sufficient number of samples in
each data category. In the bottom portion of the table the light wind speed cases (u < 6 mph)
are classified as either unstable or neutral/stable. For the high wind speed cases (u > 6 mph)
all stability classes are combined into one group since the stability is not as important in this
category. Overall the wind speed/stability classification seems to be a good manner in which
to classify the data. It is important to choose a classification that maintains an equitable
distribution of the hours across subsets. When confidence intervals are estimated for each
class, these should be based on as many data points as possible. The blocked bootstrap
resampling method is used to estimate confidence limits, as described later in this section.
The wind speed/stability classification shown in the lower portion of Table 14 was used for
the blocking criteria.
Primary Performance Measure
Both AFB (absolute fractional bias) and FB (fractional bias) are used in the comprehen-
sive evaluation. FB is used in the diagnostic evaluations, so that the tendency of a model to
underpredict or overpredict can be characterized. However, the AFB is used when
combining results for various categories or sites so that cancellation of overpredictions or
underpredictions do not occur.
77
-------
TABLE 14
TABULATION OF WIND SPEED/STABILITY
CLASSIFICATION BY SITE
Site
1
2
5
#of
Hours
142
143
75
u ^ 6 mph (2.682 m/s)
Unstable
49
77
34
Neutral
33
31
12
Stable
16
13
10
u > 6 mph
Unstable
13
9
5
Neutral
24
13
14
Stable
7
0
0
# of u ^ 6 mph
Site Hours Unstable Neutral/Stable
u > 6 mph
All stabilities
1 142 49 49
2 143 77 44
5 75-34 22
44
22
19
78
-------
When calculating either FB or AFB, the "robust highest concentration," RHC, is used
rather than the mean of the highest 25 concentrations. As discussed by Cox and Tikvart
(1990), the RHC is preferred in this type of statistical evaluation because of its stability.
Also, the bootstrap distribution of the RHCs is not artificially bounded at the maximum
predicted or observed concentration, which allows for a continuous range of concentrations.
The RHC is based on a tail exponential fit to the upper end of the distribution and is
calculated as follows
SHC = x(n) + (x -
log
(3n -
(8)
where
x
x(n)
n
average of the n-1 largest values
nth largest value
number of values exceeding the threshold value (n=26 or less)
The size of the three intersection data sets requires the value of n to be less than 26. The
value of n is nominally set to 11 so that the number of values averaged (x) is 10. A
threshold of 0.5 ppm is used.
From the overview of the data at Sites 1, 2, and 5 shown in Table 14, it appears that
several wind speed/stability class blocks can be identified and used in the diagnostic
evaluation. Within each block, a RHC is estimated for both predicted and observed
concentrations, and corresponding values of FB are formed. Therefore, several "results" are
obtained for each model. An operational evaluation based on RHC values is also made,
because the RHC for the dataset as a whole may be different than that for any of the blocks.
Furthermore, such an "overall" RHC is not a simple average of the RHC values for each
block because the RHC is not a "mean" statistic. To provide an overall assessment of model
performance, all of these "results" are factored in when creating an overall performance
measure.
Composite Performance Measure
A composite performance measure (CPM) is calculated for each model as a weighted
linear combination of the individual absolute fractional bias components. The operational
component is given a weight that is equal to the weight of the combined scientific
components. The results from the different data bases (intersections) are given equal weight.
The CPM is defined as
CPM =
(9)
79
-------
where
AFB(i)
AFB(l)
Absolute fractional bias weighted for each diagnostic
category i,
Absolute fractional bias for the operational one-hour
averages.
The wind speed (u) ^ 6 mph and neutral/stable category is weighted more than the other
two categories because of the importance of this category for regulatory modeling
purposes. Thus, the average of AFB(i) is
AVG(AFB(ff) = 0.5 AFB(u * 6 mph, NeutrallStable)
0.25 AFB(u £ 6 mph, Unstable) +
0.25 AFB(u > 6 mph, All stabilities)
(10)
Model Comparison Measures
All of the performance measures already discussed quantify the performance of one
model in reproducing the RHC observed. An ideal model will produce values of FB based on
the RHC's that are equal to zero. Any "real" evaluation will result in non-zero values. By
estimating confidence intervals for these results, we are able to quantify the significance of
these non-zero values. If the 95% confidence interval about FB for one of the models should
overlap zero, then we may conclude that the hypothesis that FB for this model equals zero
cannot be rejected with 95% confidence, so we may say that the difference from zero is not
significant. But when we compare the performance of the models, we would also like to
know if differences in their performance are significant Therefore, difference measures, such
as
FB(A) - FB(B)
(11)
are also formed and the 95% confidence intervals about them are estimated. Then we can
assess whether differences in the performance of models A and B are significant.
Differences in combined measures are also calculated. The CPM is used to calculate
pairs of differences between the models because the purpose of the analysis is to contrast the
overall performance among the models. The difference between the performance of one
model and another is the model comparison measure (MCM) defined simply as
where CPM(A)
CPM(B)
MCM(A,B) = CPM(A) - CPM(B)
Composite performance measure for Model A
Composite performance measure for Model B
(12)
80
-------
For the five models compared using the MOBILE4.1 emissions methodology, there are ten
comparison measures computed. The MCM is used to judge the statistical significance of the
apparent superiority of any one model over another.
Confidence Intervals
The bootstrap resampling technique is used to estimate confidence intervals on the
various measures described above. In applying the bootstrap procedure, observed and
predicted one-hour data are resampled for each intersection. Sampling is done with
replacement, so some hours are represented more than once. This process is repeated 1000
times so that sufficient samples are available to calculate the standard error of each measure.
At each site, the resampling recognizes the blocks selected for the diagnostic evaluation. This
assures that each of the 1000 variants of the original dataset retained the same number of
samples from each diagnostic category. Had we not blocked the data in this way, one of the
1000 variants might, for example, only consist of a few samples associated with .the largest
wind speed (repeated many times). The bootstrap resampling method allows the standard
deviation, s^, of any performance measure to be estimated, from which confidence limits can
be calculated:
95% Confidence Limits - Measure ± c sh
(13)
The standard error is simply the standard deviation of the measure over all of the bootstrap-
generated outcomes. If the measure involves a single comparison, such as FB for a single
model, then the value of c can be set equal to the student-t parameter.
Difference measures such as AFB or MCM require that simultaneous confidence intervals
be found for each pair of models in order to ensure an adequate confidence level and to
protect against falsely concluding that two models are different The method of Cleveland and
McGill (1984) is used to calculate c. In this method, c is found such that for 95 percent of
the 1,000 bootstrap i-tuples,
(14)
where Ay
= model comparison difference measure for model pair i, j,
= model comparison difference measure for model pair i, j and bootstrap
replication k, and
= standard deviation of all the A values.
For this analysis, c is found for each of the three intersections (q),
81
-------
'tfft/l
(15)
where
i andj
k
1
= model comparison difference measure (AFB or MCM) for the 1th
database, and ith and jth model,
= 1 to 5 for each model combination,
= 1 to 1000 for each bootstrap,
= intersection database, and
= 1th standard deviation of M^ for bootstrap replications 1 to 1000.
The model comparison difference measure (Myi) is based on differences in CPM and KB in
between models. Using CPM, for example, the difference in CPM values between models i
and j is calculated as
Mtjl = (CPMU -
for the primary data set, and
M
ljkl = (CPMijt/ -
for each bootstrap replication of the dataset.
(16)
(17)
Composite Model Performance Measure
The foregoing sections have identified how performance is quantified, how specific
performance measures are found, how these are combined into a single measure for each
model at each intersection, how differences in these measures between models are calculated,
and how confidence intervals are found for all of these. What remains to be done is to
calculate composite measures across all sites (intersections) used in the evaluation. A
composite model comparison measure (CM) is suggested by Cox and Tikvart (1990):
CM
(18)
where M, =
W, =
s, -
model comparison measure for the 1th data base,
1.0/S,2, and
bootstrap estimated standard error for the 1th data base.
Using the model comparison difference measure of Equation 16, bootstrap outcomes for the
composite measure can be written as
82
-------
CM,
r
M,
ijkl
tjk
(19)
With this definition, a confidence interval on CM follows that of Equation 15, so that
the value of c for FB or CPM is the value that satisfies the 95 percent criterium for
where
CM,
- CM
ijk
(20)
(21)
For each model pair, simultaneous confidence limits are placed on the composite performance
CMjj as with the 1th intersection. If the confidence limits do not overlap zero, then the
difference between the models tested is significant for these databases.
5.2.3 Summary of Scoring Scheme
In summary, the steps taken (see Section 6.2) in providing a scoring of each model
analyzed are as follows:
1) At each site, calculate the RHC for the peak one-hour observed and predicted
concentrations paired by time over all data and for each category (i.e., unstable,
neutral/stable). Calculate the FB of the RHC with confidence limits and the
AFB with confidence limits. Summarize the model performance by category.
2) Calculate the CPM for each model at each site. The smaller the CPM, the
better the overall performance of the model.
3) Calculate the MCM with confidence limits for each model pair at each site.
83
-------
4) Combine the results from all sites by calculating CM and S for each model pair
and accompanying simultaneous confidence limits,
5) Summarize the overall scores and significance of the results.
The following types of plots are presented in Section 6.2:
1) FB with confidence limits for each model as a function of the site and category
(i.e., meteorology);
2) The operational (or 1-hour) FB with confidence limits for each model as a
function of each site;
3) The operational AFB with confidence limits among the models as a function of
each site;
4) CPM for each model as a function of the site;
5) MCM with confidence limits among the models for each site; and
6) CM with confidence limits among the models over all sites.
5.2.4 Limitations of the Scoring Scheme
The traffic data available from the New York City data base are comprised of the top 50
hours of CO concentrations observed for each of three months at Sites #1 and 2 and the top
25 hours observed for each of three months at the remaining sites. This initial grouping of
the data could add a bias to the results in that it ignores situations that may have resulted in
large estimates of CO in spite of the relatively small values that were actually measured.
The large variability in the source of CO that arises due to its sensitivity to vehicle
operations tends to produce a dataset in which consecutive hours are likely to be independent
in spite of possible serial correlations in the meteorological data. Therefore, we have not
applied procedures to safeguard against improperly assessing the independence of the data
used in the evaluation. We have selected the "highest" 50 or 25 hours from each month of
the set to retain the influence of each month in the statistics. The effect of doing this has not
been assessed.
Also, the weighting used in calculating a combined score remains arbitrary. The results
of the study should be viewed in its entirety, so that conclusions reached on the basis of the
final combined measure recognize information contained in the individual measures.
84
-------
6.0 MODEL PERFORMANCE RESULTS
6.1 Phase I Results: 8 Models/6 Sites Using MOBILE4.0 Emissions
Numerous statistics, plots, and tables have been produced in order to characterize the
performance of the eight intersection modeling techniques when using MOBILE4.0 as the
emissions model. A select number of graphs and tables are presented in this subsection.
Appendix A contains a number of other analyses, including the regulatory default results, the
normal probability plots, quantile-quantile plots of the high-25 observed and predicted values,
scatterplots of observed versus predicted values using all modeled hours, and model perfor-
mance tables with all observed and predicted data paired in time/location, .time, and location
for each of the six sites analyzed.
6.1.1 Paired and Unpaired Statistics
All eight intersection modeling techniques using the MOBILE4.0 emissions were tested
at the six New York City intersections described in Section 3.0. As described in Section 5.0,
the Model Evaluation Support System (MESS), a computerized system that EPA uses for
evaluating the accuracy of air quality models, was used to generate most of the statistics.
Both paired and unpaired data sets were generated.
The average residuals (observed minus predicted CO concentration) matched by
time/location, time, and location, along with the 25-highest average unpaired values, are
presented for each site in Figures 29 through 34. These figures characterize the degree to
which each modeling technique either overestimates or underestimates the observed concen-
trations at each site. Note that the mean residual is negative when a model tends to overesti-
mate the observed concentrations. When time-paired residuals are used (whether or not
paired by location as well), the TEXIN2 model has the smallest residual values for four of the
six sites. Furthermore, all eight models indicate underpredictions when paired in time at Sites
#1 through #5. TEXIN2 and EPAINT overpredict when paired in time at Site #6. The
FHWAINT and VOL9MOB4 modeling techniques display the largest bias, on average. When
the average residuals are based on concentrations paired by location only, TEXIN2 performs
best at Sites #1, 2, and 5; GIM performs best at Site #3; CAL3QHC performs best at Site
#4; and CALINE4 performs best at Site #6. Once again, on average, FHWAINT and
VOL9MOB4 display the largest bias. The average of residuals based on the highest unpaired
25 predicted and observed concentrations indicate that no one model outperforms all other
models, although TEXIN2 displays the smallest bias at three of the sites (Sites #1, 2, and 5).
VOL9MOB4 performs best at Site #3, CAL3QHC performs best at Site #4, and CALINE4
performs best at Site #6.
85
-------
4
4>
•s
§
°s
si
8
«? o
. -a
to J
= o
§ to
2
c3
8
I
1
-------
CO O
ill '
-— CM
3 <
i
I
+^
JS
• --
e =«
2 <»
1 1
:-§
o 5
•a ^
S^
II
S "3,
•8-8
«S co
-a .is
4)
en
fi
o
87
-------
Si"
US.
c3 c
c to
O CD
*3 JZ
co en
15
co
8
.•s
„„
60 =
§
r- O »- CM
O
•3
es
1
8 |
"Jn CO
to
o
c m
il
"s|
m i
"55
CO
^ « e
•^ co 1
§i
3 •§.
o
I
bO
so
88
-------
CN
s
I03
"
_ co
3! I
IS
WJ >s
C CQ
1*8
3-5
>H
LU 2
"§
in
o
§1
S ",
§
e o
.a
"3 '
ss
T3 .
"
O
en
J3
£b
£
89
-------
«n co
n
SCM in
o
CD
•g
(8
•g
"
I
0
es
•s
**•«•
e
o
13
8
I
•a
c*
.2
•*a^
SS
O
o
S3
O
s
'aa
I
WJ
90
-------
8
CO CO
c -a
o a>
'•5 -c
co a
in '
— CO
IS
UJ CM
1®
O CD
^£ *;
^ CO
s -a
§•81
3 -i 1
•S
^s
bo
e -o
5 -S
as C/2
M «^
e =s
.2
i- O <-
-
c O
8
i
s-
« 0
•s
= <
1.1
II
o'3
> 5JO
< 2
en
a
»- O T- CM CO
91
-------
Sites #1, 2, and 5 appear qualitatively different from the other three sites in that the
relative performance of the models is independent of whether the residuals are obtained from
paired or unpaired concentrations, or whether all data are used or just the "top 25." In
contrast, the ordering of the models in terms of how near zero their bias becomes, changes at
Sites #3, 4, and 6 when concentrations are no longer paired in time. This behavior might
indicate the presence at these sites of factors that are not properly resolved in the data, or that
are not properly addressed in the model.
A second way of characterizing the performance of these models is shown in Figures 35
through 40. Here, the fractional bias between the mean predicted and the mean observed CO
concentration is labelled as the "bias of the average," while the fractional bias between the
standard deviation of the predicted concentrations and the standard deviation of the observed
concentrations is labelled as the "bias of the standard deviation." This presentation provides a
convenient means of identifying those models which produce results that are within a factor
of two of the observed values. Models with absolutely no bias in the average concentration,
and no bias in the standard deviation of the concentrations would be marked at the center of
its graph. Any symbol that lies within the central rectangle exhibits a mean and standard
deviation that is within a factor of two of those observed. With the exception of Site #3,
more models fall in the center rectangle when the top-25 concentrations are characterized,
than when all concentrations in excess of 0.5 ppm are characterized. As a group, the models
are seen to perform best at Site #6, while most tend to underestimate concentrations at Sites
#1, 2, and 5.
6.1.2 Screening Results
The screening procedure discussed in Section 5.0 has been used to characterize the
performance of the eight CO modeling techniques at six sites with the MOBILE4.0 emissions
methodology. Tables 15 through 20 present the top-25 (unpaired) observed and predicted
concentrations for each site. Included are the average and standard deviation of the top-25
predicted and observed concentrations. Also presented are the FB and the AFB of the
average and standard deviation. These values have been plotted in the right-half side of
Figures 35 through 40. In the screening procedure, emphasis is placed on those models with
a fractional bias within ± 0.67 (the factor-of-two region). That is, if both the AFB of the
average and standard deviation are ^ 0.67 (factor of two), then the models are typically
included in further evaluations. Table 21 presents a summary of those models which meet
the screening criteria for each site using MOBELE4.0 emissions. No models meet the
screening criteria at all sites. Of the three EPA intersection models (EPAINT, VOL9MOB4,
and CAL3QHC), CAL3QHC performed best. Of the two models utilizing the' FHWA
advocated average speed approach rather than explicit queuing (FHWAINT and GIM), GIM
performed best. These results were used to design the scope of phase E of this evaluation.
Therefore, the MOBELE4.1 analysis was performed for five models: CAL3QHC GIM IMM
TEXIN2, and CALINE4. '
92
-------
CO
O
§
O
O
O
CD
X
«
CM
|
"^™
£
CO
1'
m
CD
A
1
O
X
<
1
HI
1—
CO
0
X
™ l-l
M wrttr
o •
3
.
•
a
^ *
o
*
>
+
• — 1 • . . I , r
r 7
as
- (V
. ri is
p
' 1 ^ §
S ?,„ •£»
- - ! KS* a 1
w If" -3 >n *
§ hg .2 «? S
o.o o §"^
^o ^^ 4~*
^
43
*g r^ 03
- 7 si ** "M
"S -rt >^
ZBW a C "«
HOO {g sS S3
KC n •&«» 2
0* frj ^9 ^^v ^
*$ O?P
' 0!fx "IS
- N O -^ on
1 " v«^ CM
• a a|
2 §:5*
rt% lo (0 U
IH al-
-H IE g^-S
g^ g ^'w
S 4> *i»
q g rr
<1 + -S "TX
g 2 S.-C
S -S "^ ^— '
I ^ S a
- o " §SS ° -2 -2
O H O g W OS g
w §3 IS S c
H ""9 -.OO
<>•> ^ S S
r™ O O
£=* O O
H
' t
-------
CO
1
< cc
^
Q
*j
jC
CD
X
13
C\l
i
CM
CO
o
X
*
+
-
" " O
2
0.
">
0
A
CO
cc
^£
1
CM
CO
H CM
1 I
~~
~•
0 ^ S x-v .'
^" ^Ri
< i- CS g^
W '55 '
2 «J S .
I "S -&v
? M
•° g sii °-l.
0 gSH 50 «'
n > ,O ^
O • O ., O
S 1
En 0 E ^O
ZBH cn
M OO
^ 1
0 w>
a*~ E
" i
94
-------
CO
Q
o
I
(3
X
tn
CM
eo
MB
an
5 £o -is in
" -:- ill
•Oj.3
IIs 111*
a.
a.
CD
A
CO
EC
X
^
1
co
j£i
CO
OIM<<«
o
oh tnfrfo au>MrtMMO2 ^ g M
X
<
•
r ?
OOI-KW ocu tnfria a u > H >< £•< 1-1 o z
" o >n o
•3 O es
3 -CS (0
&3 ^5 ,:•
8 w ~
> O t-i
as s«g
. K W O O ^^
3S g> Qj'S'
i> ^" "S3
>. ^f V
2 o a,-^
g 5 ^-^ ^-s
> SO CO
^. ** & •*•+ t"->SC
" ta 5SI ° o o
o agg -a •=
g fag See
-------
CO
i
•<*•
I-**
ws;
a M
wo;
3 30 t
MO!
«!n
a # x
a
M. O
O
'S e
o o
'3
a S
o e
i5 8
08
en
a * x j£»
O&< U)£4O Q U > W «; t< M o 3
96
-------
H
X
H
a x
l
a * x
I
O
I
to
g
CO
W !5
l< f*
M
-------
X
D *
!§
a * x
.s«,-,
I §•'
"Q <-»
«4 4>
--o
c e
0
2 S
•S so
S
S
Q,
0.
in
o'
A
U)
1
to
co
< Ofe )f
a
+
if
SS tg
•3 + S 4J
M t*
2 43 '«3
" ? s §|I ° |
w £ 95 M c^
a So .s J=
S § -S §
g S
•-I
1
di
Si
a * x E£j
d
98
-------
TABLE 15
SCREENING TEST RESULTS FOR SITE #1
USING MOBILE4.0 EMISSIONS METHODOLOGY
Rank
1
2
3
4
5
6
7
8
9
10
11
12 '
13
14
15
16
17
18
19
20
21
22
23
24
25
AVG:
FB:
AFB:
St.Dev.
FB:
AFB:
Obs
10.6
9.6
9.1
9.0
8.7
8.6
8.4
8.3
8.2
8.2
8.2
8.0
7.8
7.6-
7.5
7.5
7.4
7.4
7.4
7.4
7.3
7.3
7.3
7.3
7.2
8.1
0.85
EPA
5.0
4.5
3.7
3.6
3.6
3.4
3.1
3.1
3.0
3.0
2.9
2.9
2.9
2.8
2.8
2.7
2.7
2.7
2.7
2.6
2.6
2.6
2.6
2.6
•2.6
3.1
0.90
0.90
0.61
0.33
0.33
FHW
2.8
2.7
2.5
2.3
2.3
2.3
2.1
2.1
2.0
2.0
2.0
1.9
1.9
1.9
1.9
1.8
1.8
1.8
1.8
1.7
1.7
1.7
1.6
1.6
1.6
2.0
1.21
1.21
0.33
0.88
0.88
V9M
4.2
3.9
3.4
2.9
2.8
2.4
2.4
2.3
2.2
2.2
2.2
2.1
1.9
1.9
1.9
1.9
1.9
1.8
1.8
1.8
1.8
1.8
1.8
1.7
1.7
2.3
1.12
1.12
0.68
0.23
0.23
C3Q
6.7
5.7
4.6
4.5
4.4
4.4
4.2
4.2
4.2
4.2
4.1
4.1
4.0
4.0
4.0
3.9
3.9
3.9
3.9
3.9
3.9
3.8
3.7
3.7
3.7
4.2
0.62
0.62
0.66
0.26
0.26
IMM
5.2
5.0
4.8
4.7
4.7
4.6
4.5
4.4
4.3
4.3
4.2
4.0
4.0
3.9
3.9
3.8
3.8
3.8
3.8
3.7
3.7
3.7
3.7
3.6
3.6
4.1 '
0.64
0.64
0.47
0.57
0.57
TEX
13.3
11.2
11.1
10.8
10.8
10.5.
10.4
10.3
10.1
9.8
9.8
9.8
9.5
9.5
9.4
9.4
9.2
9.0
8.9
8.9
8.8
8.7
8.7
8.7
8.6
9.8
-0.20
0.20
1.09
-0.24
0.24
GIM
4.3
3.8
3.6
3.4
3.4
3.3
3.3
3.3
3.2
3.2
3.1
3.1
3.1
3.0
3.0
2.9
2.8
2.8
2.8
2.8
2.8
2.7
2.7
2.7
2.7
3.1
0.88
0.88
0.39
0.74
0.74
CAL
7.3
6.8
6.8
6.7
6.7
6.6
6.1
6.0
5.9
5.9
5.9
5.8
5.7
5.6
5.5
5.5
5.5
5.4
5.3
5.2
5.1
5.1
5.1
5.1
5.0
5.8
0.32
0.32
0.66
0.26
0.26
99
-------
TABLE 16
SCREENING TEST RESULTS FOR SITE #2
USING MOBILE4.0 EMISSIONS METHODOLOGY
Rank
1
2
3
4
5
6
7
8
9
10
11
12
13
-14
15
16
17
18
19
20
21
22
23
24
25
AVG:
FB:
AFB:
St.Dev.
FB:
AFB:
Obs
11.5
10.5
10.4
10.2
10.2
9.1
8.8
8.5
8.4
8.3
8.2
8.1
8.1
8.1
8.1
8.0
8.0
7.9
7.8
7.7
7.7
7.7
7.6
7.6
7.5
8.6
1.11
EPA
6.5
6.1
6.0
5.3
5.3
5.2
5.2
5.0
4.9
4.8
4.7
4.5
4.3
4.3
4.2
4.0
3.9
3.9
3.8 .
3.7
3.7
3.6
3.6
3.6
3.6
4.5
0.61
0.61
0.86
0.25
0.25
FHW
2.3
2.2
2.2
2.0
1.9
1.8
1.7
1.7
1.6
1.6
1.6
1.6
1.6
1.5
1.5
1.5
1.5
1.4
1.4
1.4
1.4
1.4
1.4
1.4
1.4
1.6
1.36
1.36
0.28
1.20
1.20
V9M
4.3
4.0
3.7
3.7
3.6
3.3
3.3
3.2
3.1
3.0
3.0
2.9
2.9
2.9
2.9
2.8
2.8
2.7
2.6
2.6
2.6
2.5
2.5
2.5
2.4
3.0
0.95
0.95
0.50
0.75
0.75
C3Q
7.1
6.7
6.4
6.2
6.0
5.4
5.4
5.3
5.1
4.9
4.9
4.9
4.6
4.4
4.3
4.2
4.2
4.1
4.1
4.0
4.0
3.9
3.9
3.8
3.8
4. -9
0.55
0.55
0.98
0.13
0.13
IMM
5.6
5.4
5.3
5.2
5.1
5.0
4.9
4.8
4.7
4.7
4.6
4.5
4.5
4.4
4.3
4.2
4.2
4.1
4.1
4.1
4.1
4.1
4.0
4.0
4.0
4.6
0.61
0.61
0.49
0.77
0.77
TEX
10.3
10.3
10.1
9.8
8.3
7.6"
7.0
6.9
6.4
6.4
6.0
5.9
5.8
5.8
5.8
5.6
5.6
5.5
5.5
5.5
5.4
5.4
5.4
5.3
5.2
6.7
0.25
0.25
1.71
-0.43
0.43
GIM
5.1
5.0
4.9
4.9
4.9
4.3
4.1
4.0
4.0
3.9
3.9
3.9
3.9
3.9
3.8
3.8
3.8
3.8
3.8
3.8
3.8
3.7
3.6
3.6
3.6
4.1
0.71
0.71
0.48
0.79
0.79
CAL
7.7
5.8
5.0
4.8
4.7
4.7
4. 6
4. 6
4.6
4.6
4.5
4.4
4.3
4.2
4.1'
4.1
4.0
4.0
4.0
3.9
3 .9
3.9
3.9
3.7
3.6
4.5
0.63
0 . 63
0.83
0.29
0.29
100
-------
TABLE 17
SCREENING TEST RESULTS FOR SITE #3
USING MOBILE4.0 EMISSIONS METHODOLOGY
Rank
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
AVG:
FB:
AFB:
St.Dev.
FB:
AFB:
Obs
10.2
9.6
9.0
9.0
9.0
8.8
8.8
8.2
8.2
7.8
7.6
7.6
7.3
7.2
7.2
7.2
7.0
7.0
7.0
6.9
6.8
6.8
'6.8
6.7
6.7
7.8
1.02
EPA
20.1
19.6
11.0
11.0
9.9
9.4
8.4
8.3
8.3
8.3
8.2
7.9
7.8
7.5
7.5
7.5
7.3
7.1
7.1
6.8
6.7
6.6
6.6
6.6
6.5
'8.9
-0.13
0 . 13
3.53
-1.10
1.10
FHW
5.5
4.9
4.4
4.3
4.3
4.0
3.7
3.6
3.6
3.4
3.4
3.4
3.3
3.3
3.2
3.1
3.1
3.1
3.1
3.0
3.0
3.0
3.0
3.0
2.9
3.5
0.75
0.75
0.67
0.42
0.42
V9M
19.5
16.6
9.5
8.1
8.1-
7.9
7.8
7.6
7.1
7.1
7.0
6.9
6.9
6.9
6.8
6.7
6.4
6.4
6.3
6.3
6.3
6.2
6.2
5.8
5.8
7.8
-0.01
0.01
3.21
-1.03
1.03
C3Q
31.6
29.1
13.4
12.3
11.5
11.5
11.3
11.0
10.9
10.7
10.4
10.3
10.2
9.9
9.6
9.3
8.9
8.6
8.6
8.5
8.5
8.2
8.1
8.1
8.0
11.5
-0.39
0.39
5.85
-1.40
1.40
IMM
15.8
11.2
11.1
9.4
9.2
9.1
8.4-
7.9
7.9
7.9
7.8
7.5
7.2
7.2
7.2
6.7
6.5
6.4
6.3
6.0
5.8
5.6
5.5
5.3
5.2
7.8
0.00
0.00
2.34
-0.78
0.78
TEX
14.4
14.3
13.7
10.8
10.5
10.3
10.3
9.5-
9.3
9.1
8.9
8.6
8.4
8.4
8.4
7.7
7.6
7.1
7.1
7.1
7.0
6.9
6.9
6.9
6.8
9.0
-0.15
0.15
2.29
-0.77
0.77
GIM
11.9
11.3
11.1
11.1
10.3
10.2
9.3
8.9
8.9
8.8
8.5
8.5
8.1
8.0
8.0
. 8.0
7.8
7.7
7.5
7.4
7.4
7.3
7.3
7.3
7.2
8.7
-0.11
0.11
1.45
-0.34
0.34
GAL
17.3
14.4
11.5
10.7
10.3
9.9
9.9
9.9
9.6
•9.5
9.3
9.2
9.0
8.9
8.8
8.7
8.6
8.6
8.5
8.3
7.8
7.6
7.5
7.4
7.4
9.5
-0.20
0.20
2.20
-0.73
0.73
' 101
-------
TABLE 18
SCREENING TEST RESULTS FOR SHE #4
USING MOBDLE4.0 EMISSIONS METHODOLOGY
Rank
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
AVG:
FB:
AFB:
St.Dev.
FB:
AFB:
Obs
9.6
8.9
8.7
8.6
8.3
8.2
7.5
7.5
7.5
7.3
7.1
7.0
7.0
7.0
6.9
6.9
6.9
6.9
6.8
6.8
6.8
6.7
6.7
6.7
6.7
7.4
0.82
EPA
9.1
8.9
8.9
8.7
8.4
7.8
7.4
7.4
7.3
7.2
7.1
7.1
7.1
7.0
6.8
6.6
6.5
6.5
6.4
6.3
6.3
6.2
6.1
.6.0
5.6
7.1
0.03
0.03
0.99
-0.18 '
0.18
FHW
6.1
5.1
4.8
4.4
4.4
4.4
4.3
4.0
4.0
3.8
3.5
3.5
3.3
3.2
3.2
3.0
3.0
3.0
2.9
2.8
2.8
2.8
2.7
2.6
2.6
3.6
0.69
0.69
0.90
-0.09
0.09
V9M
7.3
6.7
6.7
6.6
6.5
6.1
5.9
5.8
5.8
5.3
5.3
5.3
4.9
4.9
4.9
4.8
4.8
4.8
4.7
4.6
4.6
4.6
4.6
4.5
4.4
5.4
0.32
0.32
0.85
-0.03
0.03
C3Q
12.4
11.8
11.4
10.6
10.0
9.4
9.3
8.1
7.8
7.7
7.7
7.6
7.3
7.2
7.1
7.0
7.0
7.0
6.8
6.8
6.6
6.5
6.3
6.1
5.8
8.1
-0.08
0.08
1.86
-0.77
0,77
IMM
8.0
7.9
6.5
5.4
5.3
5.1
5.0
5.0
4.9
4.9
4.4
4.4
4.3
4.1
3.9
3.9
3.8
3.8
3.8
3.5
3.5
3.4
3.4
3.4
3.3
4.6
0.47
0.47
1.29
-0.44
0.44
TEX
12.1
10.8
10.7
10.4
10.3
9.1'
9.0
8.9
8.9
8.7
8.3
8.0
7.8
7.6
7.4
7.4
7.4
7.2
7.1
' 6.5
6.4
6.4
6.2
6.2
6.1
8.2
-0.10
0.10
1.67
-0.68
0.68
GIM
12.4
11.8
11.4
10.3
10.2
9.6
9.4
9.2
9.1
8.7
8.7
7.6
7.5
7.4
7.3
7.1
7.0
6.9
6.8
6.8
6.7
6.7
6.6
6.6
6.5
8.3
-0.12
0.12
1.80
-0.74
0.74
CAL
10.3
10.3
10.0
9.8
9.6
9.3
9.3
9.2
8.7
8.6
8.3
8.2
8.2
8.0
7.9
7.9
7.8
7.5
7.5
7.4
7.3
6.9
6.8
6.6
6.5
8.3
-0.12
0.12
1.15
-0.33
0.33
102
-------
TABLE 19
SCREENING TEST RESULTS .FOR SITE #5
USING MOBILE4.0 EMISSIONS METHODOLOGY
Rank
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
AVG:
FB:
AFB:
St .Dev.
FB:
AFB:
Obs
15.5
14.6
10.4
9.9
9.6
9.3
8.9
8.7
8.6
8.6
8.4
8.4
8.0
8.0
8.0
7.9
7.6
7.6
7.5
7.5
7.4
7.4
7.4
7.4
7.2
8.8
2.07
EPA
7.6
6.0
5.8
5.4
4.8
4.6
4.4
4.3
4.0
3.8
3.5
3.4
3.3
3.2
3.2
3.2
3.2
3.2
3.2
3.2
3.1
3.1
3.1
3.0
-3.0
3.9
0.76
0.76
1.18
0.55
0.55
FHW
6.1
5.1
4.8
4.5
4.2
4.1
3.2
3.2
3.2
2.7
2.5
2.5
2.4
2.4
2.3
2.3
2.3
2.3
2.3
2.3
2.2
2.2
2.2
2.2
2.1
3.0
0.98
0.98
1.12
0.60
0.60
V9M
7.5
7.2
6.7
4.9
4.6
4.4
4.4
3.6
3.1
3.1
2.8
2.7
2.7
2.5
2.5
2.5
2.5
2.5
2.4
2.4
2.3
2.3
2.3
2.3
2.2
3.5
0.87
0.87
1.60
0.26
0.26
C3Q
10.2
7.4
6.0
5.4
4.8
4.6
4.5
4.5
4.4
4.4
4.3
4.3
4.2
4.1
4.1
4.1
4.0
4.0
3.9
3.9
3.. 9
3.8
3.8
3.8
3.7
4.6
0.62
0.62
1.41
0.38
0.38
IMM
5.4
5.1
4.7
4.5
4.4
4.2
4.2
4.1
4.1
3.9
3.8
3.8
. 3.8
3.7
3.7
3.7
3.6
3.6
3.6
3.5
3.5
3.4
3.4
3.4
3.3
3.9
0.76
0.76
0.54
1.17
1.17
TEX
10.5
9.6
9.5
9.1
8.6
8.3
7.4
7.3
6.9
6.7
6.6
6.5
6.5
6.2
6.1
6.0
6.0
6.0
5.9
5.9
5.8
5.8
5.8
5.7
5.7
7.0
0.23
0.23
1.44
0.36
0.36
GIM
8.2
7.6
7.0
5.4
5.2
4.2
3.9
3.9
3.8
3.8
3.7
3.5
3.5
3.5
3.2
3.1
3.1
3.1
3.1
3.1
3.0
3.0
3.0
2.9
2.9
4.0
0.74
0.74
1.50
0:32
0.32
CAL
8.1
8.0
7.5
7.4
7.4
7.3
6.9
6.7
6.5
6.4
6.2
6.2
6.1
6.0
5.9
5.9
5.9
5.9
5.9
5.8
5.8
5.6
5.6
5.6
5.5
6.4
0.31
0.31
0.79
0.90
0.90
103
-------
TABLE 20
SCREENING TEST RESULTS FOR SITE #6
USING MOBILE4.0 EMISSIONS METHODOLOGY
Rank
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
AVG:
FB:
AFB:
St.Dev.
FB:
AFB:
Obs
8.1
7.5
7.0
7.0
5.0
4.8
4.8
4.8
4.8
4.6
4.6
4.6
4.6
4.5
4.4
4.4
4.4
4.3
4.3
4.3
4.3
4.2
4.2
4.1
4.1
4.9
1.13
EPA
10.2
8.8
7.8
7.6
7.4
6.9
6.9
6.7
6.5
6.4
6.4
6.4
6.2
5.8
5.7
5.7
5.6
5.6
5.5
5.2
5.2
5.2
5.1
5.1
.5.1
6.4
-0.25
0.25
1.26
-0.11
0.11
FHW
9.9
8.4
7.5
5.9
5.5
5.1
4.3
4.3
4.3
4.3
4.2
4.0
3.7
3.7
3.6
3.5
3.4
3.4
3.4
3.3
3.3
3.3
3.3
3.2
3.2
4.5
0.10
0.10
1.74
-0.43
0.43
V9M
6.5
5.4
5.3
5.0
4.8
4.3
4.2
4.1
4.0
4.0
3.9
3.9
3.9
3.8.
3.7
3.7
3.7
3.6
3.6
3.6
3.6
. 3.5
3.4
3.4
3.4
4.1
0.19
0.19
0.76
0.40
0.40
C3Q
10.3
10.2
7.9
7.4
7.0
6.9
6.8
6.8
6.6
6.5
6.0
6.0
5.9
5.7
5.6
5.2
5.1
5.1
5.0
4.9
4.8
4.7
4.7
4.4
4.3
6.2
-0.22
0.22
1.58
-0.33
0.33
IMM
6.2
6.1
6.0
5.9
5.9
5.7
5.3
5.3
5.1
5.0
5.0
4.9
4.6
4.5
4.5
4.4
4.4
4.3
4.2
4.1
4.1
4.1
4.1
4.0
4.0
4.9
0.02
0.02
0.74
0.41
0.41
TEX
20.3
13.5
11.9
11.3
11.0
9.8
9.2'
9.2
9.1
8.7
8.6
8.6
8.5
8.3
8.3
8.2
8.2
8.0
8.0
7.7
7.6
7.4
7.3
7.1
7.0
9.3
-0.61
0.61
2.78
-0.84
" 0.84
GIM
9.5
8.2
7.1
5.4
5.2
5.1
5.0
5.0
4.8
4.6
4.6
4.6
4.5
4.4
4.3
4.3
4.2
4.1
4.1
4.1
4.0
4.0
4.0
4.0
3.9
4.9
0.01
0.01
1.37
-0.19
0.19
CAL
6.6
6.4
6.1
6.1
5.2
5.0
4.8
4.8
4.7
4.5
4.5
4.4
4.3
4.0
4.0
3.9
3.9
3.7
3.7
3.6
3.5
3.5
3 = 4
3.3
3.3
4.4
0.11
0.11
0.99
0.14
0.14
104
-------
TABLE 21
SUMMARY OF EPA SCREENING TEST RESULTS FOR EACH MODEL EVALUATED
IN THE NEW YORK CITY CO INTERSECTION MODELING ANALYSIS
(USING MOBILE4.0 EMISSIONS METHODOLOGY)
Site EPA
1
2 X
3
4 X
5
6 X
Total Over
All Sites 3-
Total Over
Sites #1,2,5 1
FHW V9M C3Q
X
X
X
X
XXX
124
003
IMM TEX G1M CAL
XX X
X X
X
X X
X
X XX
3 3 2 4
1302
Note: X indicates that the FB of the average and standard deviation is
within ±0.67 (factor-of-two)
105
-------
6.2 Phase H Results: 5 Models/3 Sites Using MOBILE4.1 Emissions
The MOBILE4.1 analysis was limited to the three least complex intersections with the
best quality data. When collecting and compiling the New York City database, the best
quality assurance procedures were followed at Site #1 (West/Chambers) and Site #2
(34th/8th). The uniform wind analysis summarized in Section 3.3 indicated that Sites #5
(34th/12th) and #1 are best in terms of unhindered approach wind flows and wind field
uniformity. Thus, the MOBHJE4.1 analysis was performed for the two intersections with the
best wind field uniformity (Sites #1 and #5) and one complex intersection (Site #2) which has
the best quality-assured data. As discussed in subsection 6.1.2, the MOBILE4.1 analysis
includes: CAL3QHC, GIM, IMM, TEXIN2, and CALJNE4.
6.2.1 Paired and Unpaired Statistics
All average observed and predicted CO concentrations paired in time and location and
greater than the threshold value of 0.5 ppm are presented hi Table 22. TEXIN2 exhibits the
smallest average difference (or bias) between the observed and predicted concentrations at all
three sites. At Site #1, the bias for TEXIN2 is 0.0 ppm which means there is no model bias
when the observed and predicted concentrations are paired in time and location. However,
the standard deviation of the residual values is greatest for TEXIN2 at all three sites evaluat-
ed. Furthermore, the average absolute residual and the root mean square error for TEXIN2 is
comparable with the other four models evaluated. The average absolute residual or error is
more meaningful when the bias approaches zero due to the cancellation of over- and
underpredictions. The correlation coefficient is highest for IMM at Site #1, GIM at Site #2,
and TEXIN2 at Site #5. The variance is significantly lower for the TEXIN2 model at all
three sites. For the paired comparisons, the confidence limits were estimated using the one-
sample student-t test
The highest observed and predicted CO concentrations paired in time only are presented
in Table 23. Once again, the smallest average difference between the observed and predicted
concentrations is found using TEXIN2 at all three sites. At Site #1, TEXM2 overpredicts the
highest observed concentrations by 0.7 ppm; whereas, all other models underpredict the
highest observed concentrations. At Sites #2 and 5, all models evaluated indicate an
underprediction of the highest observed concentrations. The standard deviation of the
residuals is largest for TEXIN2 at all three sites.
Table 24 presents the highest observed and predicted CO concentrations paired by station
only. All models evaluated underpredict the highest observed concentrations except for
TEXIN2 at Site #1, where TEXIN2 overpredicts by 3.9 ppm. At Site #1, CALINE4 displays
the smallest average difference or bias between the observed and predicted concentrations. At
Sites #2 and 5, the smallest bias is found using TEXIN2. The standard deviation of the
residuals is largest for TEXIN2 at Sites #1 and 2, and IMM at Site #5. The root mean square
error is largest for TEXIN2 at Site #1, CALINE4 at Site #2, and IMM at Site #5 The
106
-------
TABLE 22
ALL OBSERVED AND PREDICTED CO CONCENTRATIONS (PPM)
PAIRED IN TIME AND LOCATION USING MOBILE4.1 EMISSIONS
SITE #1
MODEL
CALINE4
CAL3QHC
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
1072
1058
1074
1058
1058
AVERAGE
OBSERVED
VALUE
3. 4
3.5
3.4
3.5
3.5
AVERAGE
DIFFERENCE
1.6
2.2
0.0
2.5
1.9
LOWER
LIMIT
1.4
2.1
-0.1
2.4
1.8
UPPER
LIMIT
1.7
2.3
0.2
2.6
2.0
STANDARD
DEV. OF
RESIDUAL
1.8
1.8
2.8
1.7
1.7
LOWER
LIMIT
• 1.7
1.7
2.7
1.6
1.6
MODEL
CALINE4
CAL3QHC
TEXIN2
GIM
IMM4
UPPER
LIMIT
1.9
1.8
3.0
1.7
1.7
ROOT
MEAN SQ
ERROR
2.4
2.8
2.8
3.0
2.5
AVERAGE
ABSOLUTE
RESIDUAL
1.9
2.3
2.1
2.5
2.1
PEARSON
CORR.
COEF.
0.457
0.406
0.400
0.456
0.479
VARIANCE
COMPARISON
1.665
2.427
0.395
5.899
2.363
LOWER
LIMIT
1.477
2.151
0.351
5.228
2.095
UPPER
LIMIT
1.877
2.738
0.445
6.655
2.666
SITE #2
MODEL
CALINE4
CAL3QHC
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
1098
1098
1098
1098
1098
• AVERAGE
OBSERVED
VALUE
3.9
3.9
3.9
3.9
3.9
AVERAGE
LOWER
DIFFERENCE LIMIT
3.0
2.8
2.4
2.9
2.7
2.9
2.7
2.3
2.8
2.6
STANDARD
UPPER DEV. OF
LIMIT RESIDUAL
3.1
2.9
2.5
3.0
2.8
1.6
1.7
1.9
1.6
1.7
LOWER
LIMIT
1.6
1.6
1.8
1.5
1.6
MODEL
C ALINE 4
CAL3QHC
TEXIN2
GIM
IMM4
UPPER
LIMIT
1.7
1.8
2.0
1.7
1.8
ROOT
MEAN SQ
ERROR
3.4
3.3
3.1
3.3
3.2
AVERAGE
ABSOLUTE
RESIDUAL
3.0
2.9
2.7
2.9
2.8
PEARSON
CORR.
COEF.
0.334
0.341
0.354
0.372
0.354
VARIANCE
COMPARISON
2.538
1.780
0.951
3.032
1.680
LOWER
LIMIT
2.255
1.581
0.845
2.693
1.492
UPPER
LIMIT
2.857
2.003
1.071
3.413
1.891
107
-------
TABLE 22 (continued)
ALL OBSERVED AND PREDICTED CO CONCENTRATIONS (PPM)
PAIRED IN TIME AND LOCATION USING MOBBLE4.1 EMISSIONS
SITE #5
MODEL
CALINE4
CAL3QHC
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
587
586
587
587
586
AVERAGE
OBSERVED
VALUE
3.8
3.8
3.8
3.8
3.8
STANDARD
AVERAGE LOWER
DIFFERENCE LIMIT
1.8
2.6
1.4
2.7
2.4
1.6
2.4
1.2
2.5
2.2
UPPER DEV. OF
LIMIT RESIDUAL
2.0
2.8
1.6
2.8
2.6
2.3
2 1
2.5
1 9
2.1
LOWER
LIMIT
2.2
2(\
2.3
1 fl
2.0
MODEL
CALINE4
CAL3QHC
TEXIN2
GIM
1MM4
UPPER
LIMIT
2.4
2.2
2.6
2.0
2.2
ROOT
MEAN SQ
ERROR
2.9
3.3
2.8
3.3
3.2
AVERAGE
ABSOLUTE
RESIDUAL
2.3
2.8
2.1
2.8
2.6
PEARSON
CORR.
. COEF .
0.246
0.232
0.277
0.259
0.182
VARIANCE
COMPARISON
1.059
1.802
0.779
3.149
2.362
LOWER
LIMIT
0.901
1.532
0.663
2.677
2.008
UPPER
LIMIT
1.245
2.119
0.916
3.703
2.778
108
-------
TABLE 23
HIGHEST OBSERVED AND PREDICTED CO CONCENTRATIONS (PPM)
EVENT BY EVENT (PAIRED IN TIME) USING MOBILE4.1 EMISSIONS
SITE#1
NUMBER AVERAGE
OF OBSERVED
MODEL EVENTS
CALINE4
CAL3QHC
TEXIN2
GIM
IMM4
142
142
142
142
142
AVERAGE
VALUE DIFFERENCE
5.8
5.8
5.8
5.8
5.8
2.0
2.8
-0.7
3.9
2.7
LOWER
LIMIT
1.8
2.6
-1.2
3.8
2.5
STANDARD
UPPER DEV. OF
LIMIT RESIDUAL
2.2
3.0
-0.1
4.1
3.0
1.4
1.3
3.2
1.1
1.3
LOWER
LIMIT
1.2
1.1
2.9
0.9
1.2
UPPER
LIMIT
1.5
1.4
3.6
1.2
1.5
SITE #2
MODEL
CALINE4
CAL3QHC
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
143
143
.143
143
143
AVERAGE
OBSERVED
VALUE
6.6
. 6.6
6.6
. 6.6
6.6
AVERAGE
DIFFERENCE
3.8
3.5
2.3
4.0
3.3
LOWER
LIMIT
3.6
3.2
1.9
3.7
3.0
UPPER
LIMIT
4.1
3.8
2.6
4.2
3.5
STANDARD
DEV. OF
RESIDUAL
1.4
1.8
2.3
1.5
1.6
LOWER
LIMIT
1.3
1.6
2.1
1.4
1.4
UPPER
LIMIT
1.6
2.0
2.6
1.7
1.8
SITE #5
MODEL
CALINE4
CAL3QHC
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
75
75
75
75
75
AVERAGE
OBSERVED
VALUE
6.2
6.2
6.2
6.2
6.2
AVERAGE
DIFFERENCE
1.6
2.7
0.8
3.9
3.3
LOWER
LIMIT
1.1
2.2
0.2
3.4
2.9
UPPER
LIMIT
2.2
3.3
1.4
4.4
3.8
STANDARD
DEV. OF
RESIDUAL
2.3
2.4
2.6
2.1
2.1
LOWER
LIMIT
2.0
2.0
2.2
1.8
1.8
UPPER
LIMIT
2.8
2.8
3.1
2.5
2.5
109
-------
TABLE 24
HIGHEST OBSERVED AND PREDICTED CO CONCENTRATIONS
PAIRED BY STATION USING MOBILE4.1 EMISSIONS
sriE#i
MODEL
CALINE4
CAL3QHC
TBXIN2
GIM
IMM4
NUMBER
OF DATA
PAIRS
8
8
8
8
8
AVERAGE
OBSERVED
VALUE
8.4
8.4
8.4
8.4
8.4
STANDARD
AVERAGE LOWER
DIFFERENCE LIMIT
2 .2
3.0
-3.9
5.0
3.1
0.4
1.4
-7.6
3.6
2.0
UPPER . DEV. OF LOWER
LIMIT RESIDUAL LIMIT
4.0
4.7
-0.3
6.4
4.3
2.0
1 8
4.1
1. 6
1-.3
1.3
1 2
2.7
1 0
0.9
MODEL
C ALINE 4
CAL3QHC
TEXIN2
GIM
IMM4
UPPER
LIMIT
4.0
3.7
8.4
3.2
2.6
ROOT
MEAN SQ
ERROR
2.9
3.5
5.5
5.2
3.3
AVERAGE
ABSOLUTE
RESIDUAL
2.3
3.0
5.0
5.0
3.1
PEARSON
CORR.
COEF.
-0.405
0.178
-0.221
-0.168
0.112
VARIANCE
COMPARISON
1.089
0.562
0.108
2.245
3.896
LOWER
LIMIT
0.218
0.113
0.022
0.449
0.780
UPPER
LIMIT
5.440
2.809
0.540
11.214
19.459
SITE #2
MODEL
CALINE4
CAL3QHC
TEXIN2
GIM
IMM4
NUMBER
OF DATA
PAIRS
8
8
8
8
8
AVERAGE
OBSERVED
VALUE
8.6
8.6
8.6
8.6
8.6
STANDARD
AVERAGE LOWER
DIFFERENCE LIMIT
5.7
4.7
2.8
5.5
4.3
3.8
2.3
-0.3
3.7
2.4
UPPER DEV. OF LOWER
LIMIT RESIDUAL LIMIT
7.5
7.2
5.9
7.2
6.2
2.1
2 7
3.5
2.0
2.2
1.4
i a
2.3
1 3 '
1.4
MODEL
CAL3QHC
TEXIN2
GIM
IMM4
UPPER
LIMIT
4.2
5.5
7.1
4.0
4.4
ROOT
MEAN SQ
ERROR
6.0
5.4
4.3
5.8
4.8
AVERAGE
ABSOLUTE
RESIDUAL
5.6
4.7
3.7
5.5
4.3
PEARSON
CORR.
COEF.
0.334
0.197
0.313
0.278
0.124
VARIANCE
COMPARISON
1.333
0.655
0.284
2.381
2.181
LOWER
LIMIT
0.267
0.131
0.057
0.477
0.437
UPPER
LIMIT
6.656
3.269
1.417
11.891
10.895
110
-------
TABLE 24 (continued)
HIGHEST OBSERVED AND PREDICTED CO CONCENTRATIONS
PAIRED BY STATION USING MOBILE4.1 EMISSIONS
SITE #5
MODEL
C ALINE 4
CAL3QHC
TEXIN2
GIM
IMM4
NUMBER
OF DATA
PAIRS
8
8
8
8
8
AVERAGE
OBSERVED
VALUE
9.4
9.4
9.4
9.4
. 9.4
AVERAGE
LOWER
DIFFERENCE LIMIT
3,1
2.8
0.5
4.0
4.5
-0.2
-0.6
-2.7
0.9
1.0
STANDARD
UPPER DEV. OF LOWER
LIMIT RESIDUAL LIMIT
6.5
6.2
3.6
7.1
8.0
3.7
3; 8
3.5
3.5
4.0
2.5
2.5
2.3
2.3
2.6
MODEL
C ALINE 4
CAL3QHC
TEXIN2
GIM
IMM4
UPPER
LIMIT
7.6
7.7
7.2
7.1
8.0
ROOT
MEAN SQ
ERROR
4.7
4.5
3.4
5.1
5.8
AVERAGE
ABSOLUTE
RESIDUAL
3.1
3.0
2.6
4.0
4.5
PEARSON
CORR.
COEF.
-0.407
0.005
0.073
0.049
-0.850
VARIANCE
COMPARISON
2.721
1.090
'1.252
1.464
4.186
LOWER
LIMIT
0.545
0.218
0.251
0.293
0.838
UPPER
LIMIT
13.589
5.444
6.252
7.313
20.907
111
-------
average absolute residual is smallest at Site #1 for CALINE4 and is smallest at Sites #2 and 5
for TEXIN2. The largest negative correlation coefficients are found using CALINE4 at Site
#1 and IMM at Site #5. At Site #2, the largest positive correlation (0.334) is found using
CALINE4.
The 25-highest observed and predicted CO concentrations over all hours and monitors
are tabulated in Table 25 for each of the sites. At Sites #1 and 2, TEXIN2 predicts peak
concentrations which exceed the peak observed values. The highest-25 averaged observed
and predicted CO concentrations, unpaired in time or location, are summarized in Table 26.
AU of the models underpredict the highest-25 concentrations at all three sites except for
TEXIN2 at Site #1. At Site #1, TEXIN2 overpredicts the 25-highest observed concentrations
by 3.9 ppm on average. CALINE4 displays the smallest bias at Site #1 and TEXIN2 displays
the smallest bias at the other two sites (only 0.5 ppm at Site #2 and 0.6 ppm at Site #5).
Furthermore, TEXIN2 has the smallest variance at all three sites. These results are similar to
the paired by station results in Table 24.
A summary of the average residual formed from predicted and observed concentrations
paired by time/location, time, and location, and the 25-highest unpaired concentrations is
shown in Figures 41 through 43. Also shown for comparison are the residuals using the
MOBILE4.0 emissions model. For most models evaluated, the residuals are lower using the
MOBELE4.1 emissions model. At Site #1, TEXIN2 overpredicts by a larger amount when
using MOBILE4.1 emissions rather than MOBILE4.0 emissions for the statistics matched by
location and the 25-highest unpaired values.' When the residuals are paired by time only,
TEXIN2 overpredicts using the MOBILE4.1 emissions; whereas, the model underpredicted
the observed concentrations using the MOBILE4.0 emissions. When comparing the two
different versions of the MOBILE emissions model, the bias is largest for TEXIN2. TEXIN2
does not include an idle correction factor. The MOBILE4.1 emissions model automatically
corrects the emissions for idle conditions. When the MOBILE4.0 version of the TEXIN2
model was applied, no idle corrections factors were applied. However, when the MOBILE4.1
version of TEXM2 was evaluated, the idle correction factors were automatically calculated by
the MOBILE4.1 emissions model causing an increase in the predicted concentrations. When
comparing the MOBILE4.0 versus the MOBILE4.1 results using the CAL3QHC model, one
should be reminded that a revised version of the model (Version 2) was tested using the
MOBILE4.1 emissions model. Further details are found in Subsection 2.2.5.
Scatterplots of all hourly observed and predicted concentrations at all receptors are
shown in Figures 44 through 46 for each model and site. The diagonal line in each plot
represents the perfect fit line. TEXIN2 displays a more even distribution of observed versus
predicted concentrations at each of the sites, especially at Site #1. However, TEXIN2 also
displays the largest overpredictions. Note that the display is limited to maximum values of
15 ppm. There are some hourly concentrations predicted by TEXIN2 that exceed 15 ppm.
The cumulative frequency distribution of the observed and predicted concentrations are
presented in Figure 47. At Site #1, TEXIN2 overpredicts the observed concentration
112
-------
TABLE 25
25 HIGHEST PREDICTED AND OBSERVED CO CONCENTRATIONS (PPM)
USING MOBILE4.1 EMISSIONS
SITE#I
Rank
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Observed
10. 6
9.6
9.1
9.0
8.7
3.6
3.4
8.3
8.2
8.2
8.2
3.0
7.8
7.6
7.5
7.5
7.4
7.4
7.4
7.4
7.3
7.3
7.3
7.3
7.2
C ALINE 4
7.9
7.3
7.3
7.2
7.2
7.0
6.4
6.4
6.3
6.2
6.2
6.2
6.1
6.0
5.9
5.9
5.9
5.8
5.6
5.6
5.5
5.5
5.3
5.3
5.3
CAL3QHC
3.1
6.3
6.2
5.4
5.1
5.0
5.0
4.9
4.3
4.8
4.7
4.7
4.7
4.6
4.6
4.5
4.4
4.4
4.4
4.2
4.2
4.2
4.2
4.2
4.2
TEXIN2
17.0
14.5
14.2
14.2
14.1
14.0
13.9
13.6
13.4
13.3
13.2
13.0
12.8
12.8
12.5
11.9
11.8
11.7
11.7
11.6
11.5
11.2
11.2
11.1
11.0
I MM
6.1
6.0
5.8
5.7
5.6
5.4
5.3
5.2
5.2
5.1
5.0
4.8
4.7
4.7
4.7
4.6
4.6
4.5
4.5
4.5
4.5
4.4
4.4
4.3
4.3
GIM
4.5
4.1
3.7
3.6
3.5
3.5
3.4
3.4
3.3
3.3
3.2
3.2
3.2
3.1
3.1
3.0
3.0
2.8
2.8
2.8
2.8
2.8
2.3
2.3
2.8
SITE #2
Rank
1
2
3
4
5
6
7
3
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
Observed
11.5
10.5
10.4
10.2
10.2
9.1
8.3
8.5
8.4
8.3
8.2
8.1
8.1
8.1
8.1
8.0
8.0
7.9
7.8
7.7
7.7
7.7
7.6
7.5
7.5
CALINE4
5.5
5.2
5.1
5.0
5.0
4.9
4.3
4.3
4.8
4.7
4.6
4.4
4.3
4.3
4.3
4.2
4.2
4.2
4.1
4.1
4.0
4.0
3.9
3.9
3.8
CAL3QHC
7.8
7.4
7.4
7.0
6.9
6.8
6.2
6.1
6.0
5.9
5.9
5.7
5.4
5.3
5.2
5.0
4.9
4.8
4.8
4.8
4.7
4.5
4.5
4.4
4.4
TEXIN2
12.8
12.5
12.4
12.2
9.6
9.1
8.4
8.4
7.8
7.6
7.6
7.2
7.2
7.2
7.0
6.9
6.6
6.5
6.5
6.4
6.3
6.3
6.3
6.3
6.2
IMM
6.3
5.9
5.4
5.4
5.3
5.2
5.1
5.0
4.9
4.9
4.9
4.8
4.3
4.8
4.3
4.7
4.7
4.7
4.7
4.7
4.7
4.6
4.6
4.6
4.5
GIM
5.3
5.2
5.2
4.9
4.8
•4.5
4.3
4.2
4.1
4.1
4.1
4.0
4.0
4.0
4.0
3.9
3.9
3.9
3.9
3.9
3.9
3.8
3.8
3.7
3.7
113
-------
TABLE 25 (continued)
25 HIGHEST PREDICTED AND OBSERVED CO CONCENTRATIONS (PPM)
USING MOBILE4.1 EMISSIONS
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
14.6
10.4
9.9
9.6
9.3
8.9
8.7
8.6
8.6
8.4
8.4
8.0
8.0
8.0
7.9
7.6
7.6
7.5
7.5
7.4
7.4
7.4
7.4
7.2
CALINE4
876
8.6
3.0
7.9
7.8
7.7
7.4
7.2
6.9
6.8
6.8
6.5
6.4
6.4
6.4
6.4
6.3
6.3
6.2
6.2
6.1
6.1
6.0
5.9
5.9
SITE #5
CAL3QHC
TT7I
9.6
7.3
7.0
5.5
5.4
5.4
5.4
5.3
5.3
5.3
5.3
5.1
5.0
5.0
4.8
4.7
4.7
4.6
4.6 .
4.6
4.6
4.5
4.5
4.5
TEXIN2
12.4
11.8
10.6
10.5
9.4
9.2
8.9
8.3
8.1
7.8
7.6
7.6
7.6
7.5
7.4
7.3
7.2
7.2
6.9
6.8
6.8
6.7
6.7
6.7
6,7
I MM
§75
6.1
5.7
5.4
5.3
5.0
5.0
4.8
4.5
4.5
4.5
4.4
4.4
4.4
4.3
4.3
4.2
4.2
4.2
4.1
4.1
4.0
4.0
4.0
3.9
GIM
8.6 ~
7 .8
7.2
5.7
5.3
4.4
4.1
4.0
4.0
3.8
3.7
3.6
3*6
3.6
3.3
3.3
3.3
3.2
3.2
3.2
3.2
3.1
3.1
3.0
3.0
114
-------
TABLE 26
25 HIGHEST OBSERVED AND PREDICTED CO CONCENTRATIONS (PPM)
UNPAIRED IN TIME OR LOCATION USING MOBILE4.1 EMISSIONS
SITE#1
MODEL
AVERAGE AVERAGE
OBSERVED PREDICTED DIFFERENCE LOWER UPPER VARIANCE LOWER UPPER
VALUE VALUE OF AVERAGES LIMIT LIMIT COMPARISON LIMIT LIMIT
CALINE4
CAL3QHC
TEXIN2
GIM
IMM4
8.1
8.1
8.1
8.1
8.1
5
4
11
3
4
.8
.5
.9
.0
.6
2
3
-3
5
3
.3
.5
.9
.0
.4
1.9
3.1
-4.5
4.7
3.0
2.7
4.0
-3.2
5.4
3 = 8
1
1
0
5
2
.512
.452
.381
.151-
.744
0.666
0.640
0.168
2.269
1.209
3.429
3.293
0.863
11.680
6.221
SITE #2
AVERAGE AVERAGE
OBSERVED PREDICTED DIFFERENCE LOWER UPPER VARIANCE LOWER UPPER
MODEL VALUE VALUE OF AVERAGES LIMIT LIMIT COMPARISON LIMIT LIMIT
CALINE4
CAL3QHC
TEXIN2
GIM
IMM4
8.6.
8.6
8.6
8.6
8.6
4
5
8
4
5
.5
.7
.1
.2
.0
4
2
0
4
3
.1
.9
.5
.4
.6
3.6
2.3
-0.5
3.9
3.1
4.6
3.5
1.5
4.8
4.1
5
1
0
5
6
.600
.114
.261
.153
.717
2.467
0.491
0.115
2.270
2.959
12.698
2.526
0.592
11.684
15.231
SITE #5
MODEL
CALINE4
CAL3QHC
TEXIN2
GIM
IMM4
AVERAGE
OBSERVED
VALUE
8.8
8.8
8.8
8.8
8.8
AVERAGE
PREDICTED
VALUE
6.8
5.6
8.1
4.2
4.6
DIFFERENCE
OF AVERAGES
2.0
3.2
0.6
4.6
4.2 .
LOWER
LIMIT
1.0
2.2
-0.4
3.6
3.3
UPPER
LIMIT
2.9
4.3
1.7
5.7
5.1
VARIANCE
COMPARISON
6.192
1.650
1.601
1.772
9.024 •
LOWER
LIMIT
2.728
0.727
0.705
0.781
3.975
UPPER
LIMIT
14.041
3.742
3.631
4.019
20.462
115
-------
CD
I
CD
CO
£
CO
CO
CD
o>
in
CM
co £
-s
X >.
JJJ g
c-<2 .
a* 23 .2
o 53 «g
Z! "c '§
•§ « o
a *•* ©
03
o
ui
=!
i
1
o
m
c
co S
s
•5
"i
x >>
IH g
I i
a S jj
2 5 -
S > |
*3 ^3
3 O C
T3 ljrS ^3
S Oi'C
> (50
< 2
8
ao
E
116
-------
CD
F
>>
CD
•a
CD
I
CM
ffl
55
1
03
a.
CD
10
C\J
I
CM
Jl
CO
3
S
8
•
^
-a
o
CS
tf
4-* sa
« -I
a s
«s «
P3 w
cu *•
•5
F
>•,
CQ
*
§
i
CM
0>
S
x >.
uj m
p
3
I
i
ffl
•a
CD
o
CO
CM
CO
O P3 SO
5 « .S
ll 3
•=f-i
rg^i -j-* *^j
i
2
as T3
3 on
•< IS <
E
117
-------
I
03
to
0
55
•a
2
"co
Q.
CD
_g>
X
tO
CVJ
10
2
55
CM T-
•
*es 09
•»-+ p~»
•?.<
••3 —
P
1
g s
.
•*- co »
3 S fi
es 3 -S
"« "O *
3 CD S
^ -g . g
o ft'C
!-> a cs
-------
• nn- • rP-
in
tn
8
CU
O
H
U
O
O
§
Ul
OQ
O
BE - in
O O 8 CJ 63 2 E-< 05
O SB I CU CU
3
03
i
CS
0.
j—J
t*HS
l*>
a
VI
r-«
•1
1
S
o
o
o
'3.
e
O,
CO
•o
u
CO 1— '
cs
s
I
5b
119
-------
u o so
«•«!
o s i o»o.s
>n
1
QO
E
120
-------
§
a.
2
o
CJ
s
a.
w
en
(Q
o
s
d
pa
i
CO
§
1
s
8
1
•3
£
a,
<»
I
o*
Is i;
S
-------
73
oa
O
I
O
O
O
s,
4>
at
3 w
^™
Co «
"a **
|3
35
2
3
E
122
-------
distribution by a large amount. All other models display underpredictions. CAL3QHC and
CALINE4 are very similar at the upper end of the cumulative frequency distribution. At Site
#2, TEXIN2 overpredicts only at the highest end of the distribution. On average, all models
underpredict the observed cumulative frequency distribution at Site #5. For this site, the
cumulative frequency distribution for TEXIN2 most closely resembles the observed cumula-
tive distribution.
The 25-highest predicted concentrations are plotted against the observed concentrations
in Figure 48. The solid, unmarked line is the 1:1 perfect fit line. As found in the cumulative
frequency distribution plots, TEXIN2 overpredicts the 25-highest observed concentrations by a
large amount at Site #1. All other models evaluated underpredict the 25-highest concentra-
tions. At Sites #2 and 5, TFJGN2 more closely follows the perfect fit line than any of the
other models evaluated. At all three sites, the next "best" model is CAL3QHC, although at
Site #1, CALINE4 predicts concentrations that are, on average, higher than the concentrations
predicted using CAL3QHC.
The fractional bias of the average (FB) is plotted versus the fractional bias of the
standard deviation (FS) for all concentrations greater than the threshold value of 0.5 ppm in
Figure 49 and for the 25-high concentrations for each model evaluated in Figure 50. All five
models evaluated are displayed for each intersection site. The center of the plot represents a
model with zero fractional bias and standard deviation or a "perfect" model. A positive value
of FB indicates that the model is underpredicting. When concentrations from all hours are
used to compute averages and standard deviations (Figure 49), nearly all models produce
standard deviations that are within a factor of two of the standard deviation of the observed
concentrations. However, only TFJON2 and CALINE4 produce averages that are within a
factor of two of the observed averages at Sites #1 and #5. Generally, all models tend to
underestimate the observed average at all three sites, with the exception of TEXIN2 at Site
#1. When only the 25-highest observed and predicted concentrations are characterized
(Figure 50), the overall bias towards underestimating the average observed concentration is
reduced. In fact all of the models except GIM produce an average concentration that is
within a factor of two of the observed concentrations at all three sites. However, the standard
deviation of predicted concentrations are more dissimilar to the standard deviation of the
observed concentrations for the high 25 concentrations.
6.2.2 Diagnostic Analysis
The statistical model evaluation results presented above are concerned only with the bias,
variance, and distribution functions of the data sets. They do not allow investigation of
whether the model is scientifically correct. For example, a model that has compensating
errors and gives right answers for the wrong reasons will still be judged "excellent" by the
statistical procedures. In order to investigate whether the models are consistent with scientific
knowledge, the model residuals (Predicted/Observed in a logarithmic system) are plotted
versus hour of the day, wind direction, wind speed, ambient temperature, Pasquill-Gifford
123
-------
o
X
t" *
\
to
X \
V
s
-3-
\ \
VL.
CM
<••*
CO
-s-
x
ui
S
55
„ 2
U 8
- 3
II
•si s
s .
X3 ^J
G ^
8 ^
g =0
O 43
&§.
| g
1-
i «
M 03
w C
43 »
-------
w u a
ai*
35
ao
tl
Ofto WC* O QMt»M«O
13
S>
125
-------
co
8 §3
rt w
u» bl
3 3
•a
•s
ana aM
ii
.3 *
> vi
I-
•-
<= ^
S3 W
CS ^J
3 §
SB
&"»•
2 S
«^
5
s
If
§ c
£ 8
so
s
126
-------
stability class, and the traffic volume. These plots are presented in Appendix B. The
residuals are grouped and plotted by means of box plots. Grouping is necessary because of
the large number of data points. The cumulative distribution function (cdf) of the residuals
within each group is represented by the 2nd, 16th, 50th, 84th, and 98th percentiles. The five
significant points in the cdf are represented using a box pattern. The number of observations
used in each box pattern are labelled near the bottom of each plot as "N = #." The solid
horizontal line represents the perfect fit and the dashed lines represent a factor-of-two.
The residuals of a good model should not exhibit any trend with model variables and should
not exhibit large deviations from unity (i.e. the residual boxes should be compact).
We would expect there to be little variation among models in patterns displayed when
the residuals are plotted against meteorological variables, because the dispersion modules are
very similar. The overall mean of the residuals for each model vary, but the underlying
pattern is indeed similar. The same is true for the variation by time-of-day, and even traffic
volume. Hence, it appears that differences among these models primarily arise in how the
emissions are determined and allocated to the various links used to describe each intersection.
In spite of there being no inter-model variations evident in these plots, it is useful to
examine the results for one model to see if any deficiencies might exist in the dispersion
modules. We have examined the results for TEXM2 with this in mind. No trends toward
increasing or decreasing residuals are seen in the plots for Site #1 (Figure B-l). At Site #2,
however, it appears that predicted concentrations tend to increase, relative to the observed
concentrations, as the atmosphere becomes more stably stratified (Figure B-6). Similarly, the
same trend can be seen by comparing the daytime residuals with those in the evening. No
such inference can be made at Site #5, possibly as a result of their being too few stably-
stratified periods to plot.
Another feature resolved by this type of plot is the present of "outliers", so long as they
occur more than 2% of the time. At Site #1, several such outliers appear in the form of very
low predicted concentrations, relative to the median value predicted in a category. The low
value seems to be associated with either the hour of day, the wind direction, or the ambient
temperature. Because the wind direction sector between 225° and 270° at Site #1 corresponds
to large buildings nearby, the flow could be disturbed, and considerably more complex, which
could lead to poorer model performance. Such an explanation is tentative, however, and more
detailed analyses would be needed to confirm such a hypothesis.
6.2.3 Regulatory Worst-Case Analysis
The ten hours with the highest observed concentrations were used to compare the
predicted concentrations using the regulatory default meteorology with the predicted concen-
trations using the observed meteorology. The comparisons for each site are presented in
Table 27. The regulatory default or worst-case meteorological conditions are defined as:
127
-------
S
11
ll3
pi
ill i
ago o
y s Q &>
ill *f
||| |
P § > w
iP
gs
S 3
sa
u a
o
rH
JJ
S:E§.
uo a
a
a
sag
BO 0,
So.
a
ii,
M a
rH
3
OH S
a
&SH
O rH O,
4-1 U Ot
a
rH
not a
5£
o a
3
01
rH
s-s
«8
s-s
§!'a
SS1
S"0
g
flfi
MU>,rocof
»n.-.mNCM,-,CMCMr«
222222S,-"-"-
a o r- p* rn
H *o n en -U- \o o o intn^>
n -7 vo
VO rt OT •» .!&
rH
Sxs
0 EH Qt
4J
n
to
So q
E u a
^ e
5 QI
ii.
Hi fit
4J
r-t
9
ns X S
o H a
&SB
S3 a
nj
rH
§.os
ao a
0) S
00.
3
CM
O
CO
in
So,
in 5
s
CM J2
« s
rH .C
SS1
g-o
So*
s5
fc
i4 0)
rs en
IIIIIIIIII
IIIIIII
ninocMCMf-tocMinm
HCM^CMrHCMCMCOrHCO
H"*""""<"<1Q
a*-Nor4in< N CM U> aicaaitnmcn
rH rH rH rH rH rH
cooococoaococoaocD
CMCMrHrHrHrHCM CM
oocnooojooomo
1
en
S
i
-
j e
u a.
. S S
OM a
r«i
o
u x s
0] frl O.
u
3 .
BOO,
3 a e
on a
Bo a
gg;
4J
rH
3
JH S
Q <
*f
t
1
SGa
CM
"
rH
o
CM JS
CM 01
S O
rH O»
CD
U u
= Q
M
O
^•oiinvicvainuivin
* rt """"""""
nnioc-uwmiomm
3:2:::-222:-
•HOOCMOrHr-irHCM-Q*.-!
•H rH tH rH *H rH rH rH tH
CMCMincMinrHGO^'COH
r-t
COCMCMCMfHCJfOr^CM
r* QO
rH
OQrHOrHOOrHino
rHrH **"
OO^fJODOIrHOOrOCJ
^inmineMrHmrHr-
amtn^rtntntno
CMrHrHrHrHrHrHrHrH
030903000)03000003
rH CMCMCMCMCMCMCM
OrHOOOOOrHO
cs"
Ui
3
S3
1
^
^ W
• •• S
^ K
?s
la
tsa
^S °*
•§ ^
A I
«N ^
-a
S "
3 a
OT" tH1
1 S
»«
u <->
t- IS
average temperalu
> = stability class
1! en
° -
™ ,O
128
-------
Wind Speed = 1 m/s
Stability Class = D
Sigma-Theta = 25° . '
Observed Temperature
"Worst Case" Wind Direction Angle (determined using ten degree increments)
Meteorological data measured at either two or three locations at each site are also listed
in the table. The peak concentrations obtained through the use of the measured meteorologi-
cal data make use of the meteorological data nearest the receptor at which the peak concentra-
tion is predicted.
As expected, the predicted concentrations found using the regulatory default meteorology
are nearly always greater than those obtained with the measured meteorology. At Site #1,
where the highest observed CO concentration unpaired in time or space is 10.6 ppm,
TEXIN2 overpredicts by more than a factor of two with a predicted concentration of 23.1
ppm. CALINE4 also overpredicts with a concentration of 13.6 ppm. CAL3QHC, with a
maximum predicted concentration of 10.4 ppm, nearly matches the maximum observed
concentration, unpaired in time or space. Using the observed meteorology, TEXIN2 also
overpredicts the maximum observed concentration with a predicted concentration of 14.5
ppm. The model with the next highest concentration (8.1 ppm) is CAL3QHC.
At Site #2, the highest observed CO concentration is 11.5 ppm. Using the default
meteorology, only TEXIN2 produces a concentration in excess of 11.5 ppm, with a maximum
predicted concentration of 13.4 ppm. The model with the next highest concentration (8.0
ppm) is CAL3QHC. Using the observed meteorology, TEXIN2 also overpredicts the
maximum observed concentration with a predicted value of 12.8 ppm. CAL3QHC, the model
with the next highest concentration, underpredicts with a concentration of 6.2 ppm.
The highest observed CO concentration at Site #5 is 15.5 ppm. As found at Site #2,
only TEXIN2 produces a maximum value (17.4 ppm) in excess of the observed peak, when
the default meteorology is used. All the other models tested underpredict the maximum
concentration, with CAL3QHC most closely matching the Tn^y-imnrn observed concentration
with a predicted value of 15.1 ppm. Using the observed meteorology, all five models
underpredict.
These results may underestimate the ability of CALINE4 and CAL3QHC to produce
peak concentrations in excess of the peak observed concentration, when the default meteoro-
logical conditions are employed. Only those hours associated with the ten highest observed
concentrations have been considered here. These hours may not coincide with those hours hi
which the peak predicted concentrations are found (using the measured meteorology). Larger
predicted concentrations can be anticipated if the entire data set is modeled with the default
meteorological assumptions.
129
-------
6.2.4 Scoring Scheme Results
The method for aggregating component results of model performance into a single
performance measure proposed by Cox and Tikvart (1990) (discussed in Section 5.2.2) was
used to compare the overall performance of the five models evaluated at three intersection
sites; The bootstrap resampling technique (Efron, 1982) was used to determine the signifi-
cance of differences in composite performance between models.
Estimates of the robust highest concentration (RHC) for one-hour averages are shown in
Table 28 along with the corresponding fractional bias for each model evaluated at the three
intersections. In general, the RHCs are largest using the operational (or entire) dataset for
each site. The FB in RHC is displayed in Figure 51 with confidence limits for each model
and site. The solid horizontal line represents a perfect fit and the dashed horizontal lines
represent a factor of two. The upper and lower values correspond to the estimated 95%
confidence limits on the fractional bias. The confidence interval is estimated from the
standard deviation of the bootstrap outcomes, and the student-t parameter. An ideal model
will produce an FB equal to zero. As shown in Figure 51, this "real" evaluation produces
non-zero values. By estimating confidence intervals for these results, the significance of these
non-zero values may be quantified. If the 95% confidence interval about the FB measure
should overlap zero, then one may conclude that there is not sufficient evidence to suggest
any model bias. In Figure 51, CALINE4 performs best at Site #1 and TEXIN2 performs best
at Sites #1 and 5. Note that only the confidence interval associated with TEXIN2 overlaps
zero.
Among the three diagnostic components in Table 28, estimates of RHC are lowest for
the higher wind speed category (> 6 mph). Each diagnostic FB category is displayed with
confidence limits in Figures 52 through 54, for each site. Of the three diagnostic categories,
the highest RHC values and best model performance are found for lower wind speeds (£ 6
mph) and neutral/stable conditions (diagnostic component 1). This category is the most
important for regulatory applications. The best value of FB in diagnostic component 1 is
found for CALINE4 at Site #1, CAL3QHC at Site #2, and TEXIN2 at Site #5.
The fractional bias in the RHC for each of the three diagnostic components can be
combined into a single measure that is called the diagnostic fractional bias. The mechanics
for doing this are similar to those discussed in Section 5.2, Equation 10. In essence, the
diagnostic FB is a weighted average of the three diagnostic components in which component
#1 (low wind speed; neutral/stable) receives a weight equal to the combined weight of the
other two components. Confidence intervals are found by means of the bootstrap technique,
and the results are presented in Figure 55. Here, CALINE4 performs best at Site #1 and
TEX3N2 performs best at Sites #2 and #5. Only the confidence intervals associated with
TEXIN2 overlap zero. A measure that includes both the diagnostic and the operational
fractional bias is constructed by averaging these two components. Again, the resulting
measure is computed for many bootstrap outcomes in order to estimate the confidence limits.
Figure 56 presents the results, which are similar to those presented earlier. The qualitative
features seen in Figures 51 (operational FB), 55 (diagnostic FB), and 56 (combined FB) are
nearly identical.
130
-------
s
CQ
CO
^*
)•*•]
CQ
J
^^
II
ll
§8
"*^ ^
jS CO
o 2 *-*
^"J O M
"5 <3 H
OS >-H SH
HP °3
^r **••>,
gg
||
§ 0
O
3 ^
HM Cfl
fe
D
§
04
i
HJ
0
M
•35
ca o
1
OCM
•HZ
U tH
OX
13 Cd
we-.
Cu
£
tH
0
a:
C"
m
s
u
a
M
IH
s
E
' M
a
c
a
3 g
2s3
£o«x •
= ~S
a fr1
0 to
o§!,
gS"
OI
rt§
" §
•o
o
s
0)
fl
§
c
0)
c
o
u
o to r- 01
CM 01 fi O
o o o o
CM m o oo
c- r* m
O O\ O P"
ca -^
fi ft
11 .0
M m
O cq
a 3
s^l^f^i;
O AJ u -U £
4J O O O
m c vi c v» c A
a> m 3 «j 3 ra a
OO Q O
cs
CO
z
J
u
3s
a o
•H
a
s
OCM
•HZ
ox
Cu
S
H
O
1
s
Ca]
tH
S
(H
c
o
«-s
gsg
rjo.x
1?
§1
° "S
*|S
»xu
™rto
2 3
1 s
"8
o
c
0)
c
o
u
OO 00 P* r-t
*o m vo oo
o o o o
o m en oo
r» vo r» vo
o o o o
S ' % * In
o o o o
1 1
in r* in m
w m in ^
0 0 0 0
oi fi p* m
o o o o
0 P* « «
^o m m
-------
SITE 1 WEST/CHflMBERS
CQ
CD
CO
U.
1 . u
1.0
0.5
0 0
-0.5
-1.0
-1 5
1 5
1.0
0.5
0 0
-0.5
-1.0
-1.5
1.5
1.0
0.5
0 0
-0.5
-1.0
-1.5
— 1 1 \ ' — 1 1
~ _ I
"If- * z
- I -
x
i t i i i
CHL3QHC IMM4 TEXIN2 GIM CHLINE4
MODEL
SITE 2 34TH/8TH
i i t i i
•••••»••-•••-»•••••••- Jr. ....... ............. JV.... ...... X. .........
' f r
1
1 1 1
CHL3QHC IMM4 TEXIN2 GIM CHLINE3
MODEL
SITE 5 33TH/12TH
i i
•"»•••*•*»• •••••«••••• ^••••••••••••« •••»•••••• ••••••••••J*a*w*a«*av,
- ., I I | .
_ _
1 1 1
CRL3QHC IMM4 TEXIN2 GIM CRLINE4
MODEL
Figure 51. The operational fractional bias (FB) with 95% confidence limits for each model as
a function of site.
132
-------
WS < 6 mph, Neutral / Stable
CO
CD
1.5
1.0 -
0.5 -
-0.5 -
-1.0 h
-1.5
i
- I
mmf
1
1
1
1
1
I
T
i
i
I
1
i
i
f
i
-
—
^
-
CF1L3QHC IMM4 TEXIN2 GIM CRLINE4
MODEL
WS < 8 mph, Unstable
1 0 »J
1.0
0.5
00
0.5
1.0
1 C
1.0
0.5
0ft
0.5
1.0
1.5
i i
.,....„..„!.
- I *
—
i i
CRL3QHC IMM4
WS>6
i i
.T..
- i I
^
1 1
1
I
1
t TEXIN2
MODEL
mph, All
i
i
i
i i
i
? "
.T-
i i
GIM CHLINE4
Stabilities
i i
I
I
I
! 1
CRL3QHC IMM4 TEXIN2 GIM CRLINE4
MODEL
Figure 52. The three diagnostic FB components with 95% confidence limits for each model at
Site #1.
133
-------
CO
CO
1.5
1.0
0.5
0.0
-0.5
-1.0
-1.5
WS < 6 mph, Neutral / Stable
......................T...................... I..........IT.....
1 I
1
CRL3QHC IMM4 TEXIN2 GIM CHLINE4
MODEL
WS < 6 mph, Unstable
•1 • *J
1.0.
0.5
0.0
-0.5.
-1.0
1.5
I — T 1 1 —
T T
••••••«.-««£«. -•••••••j.. ......... -i*. .........jr....
_
—
ill.
1
—
!
CRL3QHC IMM^ TEXIN2
MODEL
GIM CHLINE4
1.5
1.0
0.5
0.0
-0.5
-1.0
-1.5
WS > 6 mph, AH Stabilities
- f { T 1 * -
CF1L3QHC IMM4 TEXIN2 GIM CBLINE4
' MODEL
Figure 53. The three diagnostic FB components with 95% confidence limits for each model at
Site #2.
134
-------
1.5
WS < 6 mph, Neutral / Stable
1.1
0.5 *" ' v
£ 0.0
-0.5
-1.0
_1>5 I I I I I L
CRL3QHC IMM4 TEXIN2 GIM CRLINE4
MODEL
WS < 6 mph, Unstable
i.s
i i i
1.0 h I
0.5
£ 0.0
-0.5
-1.0
-1.5 I 1 1 1 1 L
CRL3QHC IMM1 TEXIN2 GIM CRLINE4
MODEL
WS > 6 mph, AH Stabilities
1.5 , , r r
1.0 I- T J
0.5 h * T" ' {'
£ 0.0
-0.5
-1.0
-1.5
CRL3QHC IMM4 TEXIN2 GIM CRLINE4
MODEL
Figure 54. The three diagnostic FB components with 95% confidence limits for each model at
Site #5.
135
-------
1.5
OlflGNOSTIC FB - SITE 1
1.0
0.5
£ 0.0
-0.5 - I
-1.0
-1.5 I 1 1 1 L
CHL3QHC IMM3 TEXIN2 GIM CHLINE4
MODEL
DlflGNOSTIC FB - SITE 2
1.5
1.0
0.5 P T f
PO 0.0 I i
-0.5
-1.0
-1.5
CRL3QHC IMM3 TEXIN2 GIM CHLINE3
MODEL
DlflGNOSTIC FB - SITE 5
'•5 I 1 1 r
1.0 h T
0.5 P T * *" i t"
s 0.01— i
-0.5
-1.0
-1.5
_L
CRL3QHC IMM4 TEXIN2 GIM CRLINE4
MODEL
Figure 55. The combined diagnostic FB with 95% confidence limits for each model as a
function of site.
136
-------
COMBINED F8 - SITE 1
1.5
1
I
1.0
0.5
-0-5 ^ i.
-1.0
.1.5 I 1 1 1 i i
CFIL3QHC IMM4 TEXIN2 GIM CHLINE4
MODEL
COMBINED FB - SITE 2
1.5 , , , , ,
1.0 -
0.5 - "j ? * 5
CO 0.0 _, ^
-0.5 -
-1.0 -
-l.S I 1 1 1 1 1
CHL3QHC IMM1 TEXIN2 GIM CHLINE4
MODEL
COMBINED FB - SITE 5
1-5 • | | | | -r
1.0
a.sk I *• •••- i f
CD (/Id I T
^_ 10.10 I j-
-0.5
-1.0
-1.5
i
i
i
i
i
CBL3QHC IMM4 TEXIN2 GIM CflLI.NE
-------
Because the fractional bias is a signed measure, small values may be obtained when
several values with opposite signs are combined. In characterizing the overall performance of
these models, we also wish to use a measure that avoids such "cancellation" effects. This is
accomplished by using the absolute fractional bias, AFB. In particular, a composite perfor-
mance measure (CPM) for the models is formed as a weighted linear combination of the
individual absolute fractional bias components (see Equations 9 and 10 in Section 5.2.2). The
CPM values with 95% confidence limits are presented in Figure 57. The smaller the CPM
value, the better the overall performance of the model. Results for CAL3QHC, IMM, GIM,
and CALJNE4 are the same as those found for the combined FB in Figure 56. This is due to
the fact that none of the models produce an FB that is negative, so there is no difference
between AFB and FB. TEXIN2 does produce FBs of both sign, so the characteristics of its
AFB are different from those of its FB. At Site #2 for example, FB for TEXIN2 is not
significantly different from zero, whereas the corresponding CPM is significantly different
from zero.
A^ further combination is made in order to construct a performance measure across all
three sites. This is the composite model comparison measure (CM), which is made up of the
CPM values calculated at each site (see Equation 18 in Section 5.2.2). The results, shown in
Figure 58, indicate that the best performing models are TEXIN2, CALINE4, and CAL3QHC,
with TEX3N2 having the lowest overall CM value using the CPM statistics. Similarly, the
AFB from diagnostic category 1 (u ^ 6 mph, neutral/stable) can also be combined over all
three sites into a single CM. As shown in Figure 59, CAL3QHC has the lowest CM by a
factor of two from the next best model (TEXIN2). As mentioned earlier, this category is
typically most important in terms of regulatory applications.
So far, all of the performance measures presented in this section quantify the perfor-
mance of each model in reproducing the RHC that is observed. When comparing these
performance measures, one would like to know if differences are significant. Therefore, as
discussed in Section 5.2.2, two separate difference measures have been formed and the 95%
confidence intervals about them have been estimated. The first difference measure evaluated
is DFB or the difference in FB between two models denoted as A and B (AFB(A,B)).
Figures 60 through 62 present the AFB or DFB statistics with 95% confidence limits for each
pair of models for each site. As discussed in subsection 5.2, the method of Cleveland and
McGiU (1984) was used to calculate simultaneous confidence intervals for each pair of
models in order to ensure an adequate confidence level and to protect against falsely
concluding that two models are different. Using the AFB statistics, it may be concluded that
at Site #1, TEXIN2 is significantly different from the other models. However, at Site #2
TEXIN2 is not significantly different from CAL3QHC, and it is not significantly different
from either CAL3QHC or CALINE4 at Site #5.
Differences in combined measures were also calculated. As discussed in Section 5.2.2,
the CPM or composite performance measure was used to calculate pairs of differences
between the models because the purpose of the analysis is to contrast the overall performance
among the models. The difference in CPM values between one model and another model is
the model comparison measure (MCM). The MCM is used to judge the statistical signifi-
cance of the apparent superiority of any one model over another. The MCM statistics with
138
-------
SITE 1 WEST/CHflMBERS
0_
u
Q_
C_J
1 »4.
1.0
0.3
0.S
0.4
0.2
0 0
1 1 l .
- -
T
I I { [ -
1 ,,
i i
CRL3QHC IMM4 TEXIN2 GIM CRL1NE4
MODEL
SITE 2 34TH/8TH
1.2
1.0
0.3
0.6
0.4
0.2
0 0
i i
- -
: ! I I :
: i { :
l i
CRL3QHC IMM4 TEXIN2 GIM CRLINE4
MODEL
SITE 5 34TH/12TH
l °
1.0
0.8
0.4
0.2
0.0
iii-
— -
~~ 1 1 ~"
1 1
— ,, •'• -
i
CflLSQHC IMM4 TEXIN2 GIM CRLINE4
MODEL
Figure 57. The composite performance measure (CPM) with 95% confidence limits for each
model as a function of site.
139
-------
CFM STHTISTICS
C_J
1.0
0.8
a.s
0.1
0.2
0.0
I I
I -
CHL3QHC IMM<5 TEXIN2 GIM CHLINE4
MODEL
Figure 58. The composite model comparison measure (CM) with 95% confidence limii
CPM statistics.
ts using
140
-------
1.0
0.8
0.6
0.4
0.2
0.0
RFB OF DIflGNOSTIC CATEGORY 1
CPL3QHC IMM4 TEXIN2 GIM CRLINE4
MODEL
Figure 59. The composite model comparison measure (CM) with 95% confidence limits using
the AFB of dmgnostic category 1 ( U <; 6mph, neutral/stable ) statistics.
141
-------
CO
cu
- i
on
I
I
O\
n m v> m ** m u
- - - - - r -
&a
'
i
i
•
.
.
tuisj -now ssnu -ran*
to an xu mi on TO HIO xu mi on
CQ
I
03
g
so
142
-------
w
aa
tf
X
3
I
•a
0,
"o
•o
o
1
S
I
d
3
s^
§
S
•a
« m
tun
0\
^
•^^
ST
Q
a
^
VO
§
so
143
-------
5
*
*
o
M
5
•
m m
i m
'
1 V
w «
• <
fl
.
M •
•2 •
a
a
jl
* ?
a
• n
* T
a
• U
7
• u
7 ^
• «
7
•
;
"B
£ *
t
J
y
»I
|i
X
mtuixm
[9 XU M
d
5
" 1
o
03
li
o
•S
"i
1
I
CQ
IX,
Q
•^s
CQ
-------
simultaneous 95% confidence limits for each model pair are presented in Figures 63 through
65 for each site. Since TEXIN2 had the best composite model performance measure when
using the CPM statistics, it is important to test if this model is significantly different from all
other models evaluated. As shown in Figure 63, at Site #1, TEXIN2 is not significantly
different from CAL3QHC, IMM, and GIM. At Site #2 (see Figure 64), TFJON2 is not
significantly different from CAL3QHC and IMM. At Site #5 (see Figure 65), TFJON2 is not
significantly different from CALINE4. Furthermore, when the MCM statistics from each site
are combined into one composite model comparison measure (CM), TEXIN2 is not signifi-
cantly difference from either CAL3QHC or CALINE4 (see Figure 66). A summary of the
CM statistics including the standard error (S) and the ratio of CM to S is presented in Table
29 for each pair of models. Also included in Table 29 is the composite value of c for CPM
over each site that ensures an adequate confidence level and protects against falsely conclud-
ing that two models are different. If the ratio of CM/S is greater than ±c, then it may be
concluded with 95% confidence that the two models are significantly different As shown in
Table 29, the following model pairs are not significantly different with 95% confidence:
CAL3QEKyTEXIN2 and TEXIN2/CALINE4.
145
-------
ri
3 i
0
5 :
(D
1
J3
£
a
**•«
s
-------
m
KM
1
o
o
S
§
1
S
1
va
CB
1
3 cs
r2 a
t- Q.
1
ao
£
147
-------
CRIIE1
*
C3
H
S
i
«
i
V
"""2"-
- • « T 7 7
IQM
»•!••» • M
= - -i • i 7 7
KM
-
• • M « n « M
: • - .• m j - -
KM
•
-
» • t* • M • ul
: - - - 7 r 7
KM
'
•
i
1
-------
3
8?
ri
£
CO
^•l
•g
M4
O
3
^
to
rf
m
n
3
s
5 :
3
-------
TABLE 29
SUMMARY OF THE COMPOSITE MODEL COMPARISON MEASURES (CM)
OF DIFFERENCES BETWEEN MODEL PERFORMANCE AS MEASURED BY THE
ABSOLUTE FRACTIONAL BIAS IN PREDICTING ROBUST HIGHEST
CONCENTRATIONS FOR MCM STATISTICS
Model 1
Model 2
CM
CM/S
CAL3QHC
CAL3QHC
CAL3QHC
CA13QHC
IMM4
IMM4
IMM4
TEXIN2
TEXIN2
GIM
IMM4
TEXIN2
GIM
CALINE4
TEXIN2
GIM
CALINE4
GIM
CALINE4
CALINE4
-0.226
0.087
-0.374
-0.137
0.316
-0.166
0.132
-0.432
-0.092
0.245
0.044
0.067
0.044
0.043
0.067
0.041
0.035
0.066
0.063
0.041
2.99
2.99
2.99
2.99
2.99
2.99
2.99
2.99
2.99
2.99
-5.07
1.29
-8 .59
-3.19
4.74
-4.02
3.80
-6.54
-1.45
5.94
Notes: S = standard error in CM based on bootstrap outcomes.
c =s factor for the 95% simultaneous confidence interval.
The hypothesis that there is no difference in model performance can be rejected with
95% confidence if | CM/S I > c.
150
-------
7.0 SUMMARY AND CONCLUSIONS
An evaluation of performance of eight modeling techniques (CAL3QHC, FHWAINT,
GIM, EPAINT, CALINE4, VOL9MOB4, TEXIN2, and IMM) in simulating concentrations of
CO at the six intersections monitored as part of the Route 9A Reconstruction Project in New
York City is presented in this report. A phase I study evaluated the performance of all eight
modeling techniques at all six intersections. Estimates of the emission rate of CO were
obtained from MOBILE4.0. A new version of this model, MOBILE4.1, was released as the
phase I study was completed. Results obtained during phase I were used to identify a subset
of modeling techniques and intersections for a phase n study in which estimates of the
emission rate of CO were obtained from MOBILE4.1. Of the three EPA intersection models
evaluated in Phase I (EPAINT, VOL9MOB4, and CAL3QHC), CAL3QHC performed best.
Of the two models utilizing the FHWA advocated average speed approach rather than explicit
queuing (FHWAINT and GIM), GIM performed better. Therefore, the phase H study with
MOBILE4.1 was performed for five models: CAL3QHC, GIM, IMM, TEXIN2, and
CALINE4. A uniform wind analysis conducted for each site indicated that Sites #5
(34th/12th) and #1 (West/Chambers) are best in terms of unhindered approach wind flows and
wind field uniformity. The best quality assurance procedures were followed at Sites #1 and
#2 (34th/8th) when collecting and compiling the New York City database. Therefore, the
phase n MOBILE4.1 study was performed for three intersections (Sites #1, 2, and 5).
Two types of statistical evaluations of differences between observed and modeled CO
concentrations are performed. First, the EPA Model Evaluation Support System (MESS) is
used to calculate a standard set of performance measures and statistical estimators. Second, a
scoring scheme is used to aggregate the component results of model performance into a single
performance measure- used to compare the overall performance of the models and the boot-
strap resampling technique is used to determine the significance of differences in composite
performance between models.
The phase I results using eight models with MOBILE4.0 emissions at six sites indicate
that, on average, FHWAINT and VOL9MOB4 display the largest bias. When time-paired
comparisons are made, TEXIN2 has the bias nearest zero at four of the six sites and all eight
modeling techniques indicate underpredictions at Sites #1 through #5. When the average
residuals are based on concentrations paired by location only, TEXIN2 performs best at Sites
#1, 2, and 5; GIM performs best at Site #3, CAL3QHC performs best at Site #4; and
CALINE4 performs best at Site #6. The average residuals based on the highest unpaired 25
predicted and observed concentrations indicate that no one model consistently outperforms all
other models. In fact, when the fractional bias of the mean and standard deviation of the
highest 25 predicted CO concentrations relative to the mean and standard deviation of the 25
highest observed concentrations is used as an indicator of performance, none of the models
produces fractional biases less than or equal to 0.67 in absolute value across all of the sites.
Sites #1, 2, and 5 appear qualitatively different from the other three sites in that the
relative performance of the models is independent of whether the residuals are obtained from
151
-------
paired or unpaired concentrations, or whether all data are used or just the "top 25." In
contrast, the ordering of the models in terms of how near zero their bias becomes, changes at
Sites #3, 4, and 6 when concentrations are no longer paired in time. This behavior might
indicate the presence at these sites of factors that are not properly resolved in the data, or that
are not properly addressed in the model. Recall that Sites #1, 2, and 5 are the least complex
sites in the group.
The phase IE study indicates that the performance of the five models when MOBILE4.1
is used is qualitatively similar to the performance seen in phase I when MOBELE4.0 is used.
Key points discovered in this evaluation include the following:
1. Effect of Using MOBILE4.1 Relative to MOBILE4.Q
Larger CO concentrations are generally predicted by all models at all sites. TEXIN2
exhibits the greatest change in bias, apparently due to the use of correction factors to
emissions during idle conditions.
2. Mean Bias Exhibited by Each Model
When the observed and predicted concentrations are paired in time and location,
TEXIN2 exhibits the smallest average bias at all three sites. For the paired in time
only residuals, all models underpredict the highest observed concentrations at all
three sites, except TEXIN2 at Site #1. When paired by station only, CALINE4
displays the smallest average bias at Site #1 and TEXIN2 displays the smallest bias
at Sites #2 and #5. All of the models underpredict the highest-25 concentrations at
all three sites except for TEXIN2 at Site #1.
3. Influence of Meteorology on Model Performance
CAL3QHC performs better at lower wind speeds. Relative performance among the
other models does not change in a consistent manner as a function of the .meteorolo-
gy. Hence, differences between models are primarily related to how emissions are
determined and allocated to the links used to describe each intersection.
4. Model Performance with "Regulatory Default Meteorology"
The predicted concentrations found using the regulatory default meteorology are
nearly always greater than those obtained with the measured meteorology. At Site
#1, TEXIN2 overpredicts the maximum observed concentration by more than a factor
of two; whereas, CAL3QHC nearly matches the maximum observed concentration,
unpaired in time or space. At Site #2, TEXIN2 overpredicts while CAL3QHC, the
next highest modeled concentration, underpredicts. At Site #5, TEXIN2 also
overpredicts the maximum observed concentration, while CAL3QHC nearly matches
it.
152
-------
5. Fractional Bias of the Robust Highest Concentrations (RHQ
The analysis of the fractional bias (FB) of the robust highest concentration (RHC) for
all one-hour averages indicates that CALINE4 performs best at Site #1, and TEXIN2
performs best at Sites #2 and 5. The confidence interval associated with TEXIN2
indicates that its FB is not significantly different from zero at the 95% confidence
level at Sites #2 and 5.
6. Diagnostic Evaluation of Performance for the RHC
When performance is evaluated for three diagnostic categories, the best model
performance is found for light wind and neutral/stable conditions. This category is
most important for regulatory applications. The best value of FB for this diagnostic
category is found for CALINE4 at Site #1, CAL3QHC at Site #2, and TEXIN2 at
Site #5. When results for these sites are combined by forming a comparison
measure (CM) based on the absolute FB(AFB), the CM indicates that CAL3QHC
performs best for the category containing light winds and neutral/stable dispersion.
7. Overall Evaluation of AFB in RHC
An analysis of the weighted linear combination of the individual AFB components
(operational and diagnostic) or composite performance measure (CPM) indicates that
when the results from all three sites are combined into one composite model perfor-
mance measure (CM), TEXIN2 performs best and the performance of CAL3QHC is
very close to that of TEXIN2.
8. Significance of Performance Results
The model comparison measure (MCM), which is the difference in CPM values
between one model and another model, is used to judge the statistical significance of
the apparent superiority of any one model over another. Since TEXIN2 has the best
composite model performance measure when using the CPM statistics, it is important
to test if its performance is significantly different from all other models evaluated.
At Site #1, TEXIN2 is not significantly different from CAL3QHC, IMM, and GIM.
At Site #2, TEXM2 is not significantly different from CAL3QHC and IMM. At Site
#5, TEXEST2 is not significantly different from CALINE4. When the MCM statistics
from each intersection are combined, TEXIN2 is not significantly different from
either CAL3QHC or CALINE4 with 95% confidence.
153
-------
-------
8.0 REFERENCES
Benson, P., 1979: CALINE 3 - A Versatile Dispersion Model for Predicting Air Pollutants
Levels Near Highways and Arterial Streets. Report No. FHWA/CA/TL-79/23. Office of
Transportation Laboratory, Sacramento, CA.
Benson, P., 1989: CALINE4 - A Dispersion Model for Predicting Air Pollutant Concentra-
tions Near Roadways. Report No. FHWA/CA/TL-84/15. Office of Transportation
Laboratory, Sacramento, CA.
Benson, P., 1991: Personal Communication. April 8, 1991.
Bullin, G., J. Korpics, and M. Hlavinka, 1990: User's Guide to the TEXIN2/MOBILE4
Model. Research Report 283-2. Texas State Department of Highways and Public Trans-
portation. College Station, TX.
Cleveland, W.S. and R. McGill, 1984: Graphical Perception: Theory, Experimentation, and
Application to the Development of Graphical Methods. /. Am. Stat. Assoc., 79, 531-554.
Conway, R.F. and J. Zamurs, 1991: A Technique for Improving Carbon Monoxide Intersec-
tion Air Quality Model Performance. 84th AWMA Annual Mtg., Vancouver, B.C.
Cox, W.M., 1988: Protocol for Determining the Best Performing ModeL U.S. EPA, OAQPS,
Technical Support Division, Source Receptor Analysis Branch. Research Triangle Park,
North Carolina.
Cox, W.M. and J.A. Tikvart, 1990: A Statistical Procedure for Determining the Best Per-
forming Air Quality Simulation Model. Atm. Env., 24, 2387-2395.
DiCristofaro, D., R. Yamartino, and R. Mentzer, 1991: Development of New York City
Database and Protocol for Evaluation of Intersection Modeling Techniques: Tasks 1 and
3 Results. Sigma Research Corporation, WA 2-1.
Efron, B., 1982: The Jackknife, the Bootstrap and Other Resampling Plans. Society for
Industrial and Applied Mathematics, Philadelphia,. PA.
EMI Consultants, 1985: The Georgia Intersection Model for Air Quality Analysis.
Knoxville, TN.
ENSR, 1988: Monitoring and Quality Assurance Plans for the Route 9A Reconstruction
Carbon Monoxide and Meteorological Program. Doc. 7082-001-026. ENSR Consulting
Co., Acton, MA.
EPA, 1978: Carbon Monoxide Hot Spot Guidelines, Volumes I-V. EPA-450/3-78-033
through EPA-450/3-78-037. Research Triangle Park, North Carolina.
155
-------
EPA, 1979: Guidelines for Air Quality Maintenance Planning and Analysis, Volume 9
(Revised): Evaluating Indirect Sources. EPA-450/4-78-001. Research Triangle Park
NC
EPA, 1987a: Ambient Monitoring Guidelines for Prevention of Significant Deterioration
(PSD). EPA-450/4-87-007. Research Triangle Park, NC.
EPA, 1987: Model Evaluation Support System (MESS) Documentation. EPA-450/4-87-004.
Research Triangle Park, NC.
EPA, 1989: User's Guide to MOBILE-4 (Mobile Source Emissions Model). EPA-AA-TEB-
89-01. Ann Arbor, ML
EPA, 1991: User's Guide to MOBILE4.1 (MOBILE Source Emission Factor Model). EPA-
AA-TEB-91-01. Ann Arbor, MI.
EPA, 1992: User's Guide to CAL3QHC Version 2.0: A Modeling Methodology for Pre
dieting Pollutant Concentrations Near Roadway Intersections. EPA-454/R-92-006. U.S.
Environmental Protection Agency, Research Triangle Park, NC.
Fox, D., 1981: Judging Air Quality Model Performance (A Summary of the AMS Workshop
on Dispersion Model Performance, Woods Hole, MA, 8-11 September 1980). Bull. Am.
Meteorol. Soc., 62, 599-609.
Kunselman, P., 1974: Automobile Exhaust Emission Modal Analysis Model. EPA-460/3-74-
005. U.S. Environmental Protection Agency. Ann Arbor, ML
Nudelman, H., 1991: Personal Communication. November 18, 1991.
NYDOT, 1982: Intersection Midblock Model User's Guide. New York State Department
of Transportation, Albany, NY.
PEI, 1988: Development and Review of Traffic and CO Emission Components of Intersec-
tion Modeling Techniques. U.S. Environmental Protection Agency, Research Triangle
Park, NC.
Transportation Research Board (TRB), 1985: The Highway Capacity Manual: Special Report
209. Washington, D.C.
156
-------
APPENDIX A
ADDITIONAL PHASE I MOBILE4.0 ANALYSES
-------
-------
TABLE A-l
ALL OBSERVED AND PREDICTED CO CONCENTRATIONS (PPM)
PAIRED IN TIME AND LOCATION USING MOBILE4.0
SITE#1
MODEL
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
MODEL
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
1053
1069
1056
1051
1071
1058
1058
UPPER
LIMIT
1.8
1.8
1.8
1.8
2.4
1.7
1.7
AVERAGE
OBSERVED AVERAGE LOWER UPPER
VALUE DIFFERENCE LIMIT LIMIT
3.5
3.5
3.4
3.5
3.5
3.4
3.5
3.5
ROOT
MEAN SQ
ERROR
3.1
3.3
2.4
2.9
3.4
2.4
3.0
2.7
2.9 2.8 3.0
1.7 1.6 1.3
2.4 2.3 2 5
2.9 2.8 3.0
0.7 0.6 0.9
2.5 2.4 2.6
2.1 2.0 2.2
AVERAGE PEARSON
STANDARD
DEV. OF
RESIDUAL
1.7
1 7
1 7
1 7
1.7
2.3
1.7
1.6
ABSOLUTE CORR. VARIANCE LOWER
RESIDUAL COEF. COMPARISON LIMIT
2.6 0.413 6.427
2.9 . 0.467 16.386
1.9 0.455 1.875
2.5 0.420 3.275
2.9 0.382 13.432
1.9 0.409 0.670
2.5 0.453 6.307
2.3 0.482 3.369
5.696
14.519
1.663
2.902
11.901
0.594
5.590
2.986
LOWER
LIMIT
1.6 — '
1 fi
I 7
1 K
1 6
2 2
1 6
1.6 •
UPPER
LIMIT
7.251 • "•
18.492
2.114
3.695
15.160
0.755
7.116
3.801
SITE #2
MODEL
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
MODEL
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
1098
1098
1098
1098
1098
1098
1098
UPPER
LIMIT
1.7
1.7
1.7
1.6
1.8
1.6
1.7
AVERAG
OBSERVED
VALUE
3.9
3.9
3.9
3.9
3.9
3.9
3.9
3.9
ROOT
MEAN SQ
ERROR
3.4
3.9
3.4
3.4
3.6
3.2
3.3
3.3
AVERAGE LOWER ' UPPER
DIFFERENCE LIMIT LIMIT
3.0 2.9 3.1"'
3.5 3.4 3.6
3.0 2.9 3.1
3.0 2.9 3.1
3.2 3.1 3.3
2.7 2.5 2.8
2.9 2.8 3.0
2.8 2.7 2.9
AVERAGE PEARSON
ABSOLUTE CORR . VARIANCE
RESIDUAL COEF. COMPARISON
3.0 0.355 2.993
3.5 0.309 22.750
3.0 0.330 2.702
3.0 0.332 2.431
3.2 0.363 6.163
2.8 0.342 1.422
2.9 0.370 3.238
2.8 0.340 2.089
STANDARD
DEV. OF
RESIDUAL
1.6
1 6
1 6
1.7
1. 6
1 8
1. 6
1.7
LOWER
LIMIT
2.659
20.209
2.401
2.160
5.475
1.263
2.877
1.856
LOWER
LIMIT
— m
1 5
1 £
1.6
1 5
1 7
1 5
1.6
UPPER
LIMIT
3.369
25.610
3.042
2.737
6.938
1.601
3.645
2.351
A-i
-------
TABLE A-l (continued)
ALL OBSERVED AND PREDICTED CO CONCENTRATIONS (PPM)
PAIRED IN TIME AND LOCATION USING MOBILE 4.0
SITE #3
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
481
479
483
479
478
480
481
481
AVERAGE
OBSERVED AVERAGE
VALUE
3.8
3.8
3.7
3.8
3.8
3.8
3.8
3.8
STANDARD
LOWER UPPER DEV. OF
DIFFERENCE LIMIT LIMIT RESIDUAL
1.
2.
0.
1.
1.
1.
1.
1.
3
7
8
4
9
6
2
6
1
2
0
1
1
1
0
1
.2
.5
.5
.1
.7
.4
.9
.3
1.7
2.9
1.0
1.8
2.2
1.9
1.4
1.8
2.9
2.1
3.1
3.5
2.7
3.0
2.8
2.6
LOWER
LIMIT
2.7
2.0
2.9
3.3
2.5
2.8
2.7
2.4
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
UPPER
LIMIT
3.1
2.2
3.3
3.8
2.9
3.2
3.0
2.8
ROOT
MEAN SQ
ERROR
3.2
3.4
3.2
3.8
3.3
3.4
3.1
3.0
AVERAGE
ABSOLUTE
RESIDUAL
2.6
2.9
2.5
2.9
2.7
2.8
2.5
2.5
PEARSON
CORR.
COEF.
-0.011
-0.019
-0.088
-0.079
-0.016
-0.007
0.009
-0 . 022
VARIANCE
COMPARISON
0
4
0
0
0
0
0
1
.800
.640
.721
.439
.989
.696
.807
.205
LOWER
LIMIT
0.669
3.877
0.603
0.367
0.826
0.582
0.675
1.007
UPPER
LIMIT
0
5
0
0
1
0
0
1
.957
.552
.362
.525
.183
.832
.966
.441
SITE #4
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
436
436
437
436
436
437
436
436
AVERAGE
OBSERVED
VALUE
3.9
3.9
3.9
3.9
3.9
3.9
3.9
3.9
AVERAGE LOWER
DIFFERENCE LIMIT
2.0
2.9
1.6
2.0
2.4
1.6
1.6
2.4
1.7
2.7
1.4
1.7
2.2
1.4
1.3
2.2
STANDARD
UPPER DEV. OF
LIMIT RESIDUAL
2.2
3.0
1.9
2.2
2.6
1.9
1.8
2.6
2.4
1.8
2.6
2.5
2.1
2.6
2.6
2.0
LOWER
LIMIT
2.2
1.7
2.4
2.4
1.9
2.5
2.5
1.9
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
UPPER
LIMIT
2.5
2.0
2.8
2.7
2.2
2.8
2.8
2.2
ROOT
MEAN SQ
ERROR
3.1
3.4
3.0
3.2
3.2
3.1
3.1
3.1
AVERAGE
ABSOLUTE
RESIDUAL
2.6
2.9
2.5
2.7
2.7
2.6
2.5
2.6
PEARSON
CORR.
COEF.
0.056
0.081
0.060
0.055
0.064
0.073
0.059
0.014
VARIANCE
COMPARISON
0.874
3.555
0.649
0.681
1.523
0.602
0.602
2.083
LOWER
LIMIT
0.724
2.945
0.538
0.564
1.261
0.499
0.499
1.726
UPPER
LIMIT
1.055
4.291
0.784
0.822
1.338
0.726
0.727
2.515
A-2
-------
TABLE A-l (continued)
ALL OBSERVED AND PREDICTED CO CONCENTRATIONS (PPM)
PAIRED IN TIME AND LOCATION USING MOBILE 4.0
SITE #5
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
586
585
587
586
585
587
587
586
AVERAGE
OBSERVED
VALUE
3.3
3.8
3.8
3.8
3.3
3.8
3.8
3.8
AVERAGE LOWER UPPER
DIFFERENCE LIMIT LIMIT
2.8
3.0
1.9
2.8
3.1
1.7
2.7
2.6
2.7 3.0
2.9 3.2
1.7 2.1
2.6 2.9
2.9 3.2
1.5
2.6
2.4
1.9
2.9
2.7
STANDARD
DEV. OF
RESIDUAL
1.9
1.8
2.2
2.0
1.9
2.3
1.9
2.0
LOWER
LIMIT
1.8
1.7
2.1
1.9
1.8
2.1
1.8
1.9
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
UPPER
LIMIT
2.0
1.9
2.4
2.1
2.0
2.4
2.0
2.1
ROOT
MEAN SQ
ERROR
3.4
3.5
2.9
3.4
3.6
2.8
3.3
3.3
AVERAGE
ABSOLUTE
RESIDUAL
2.9
3.0
2.3
2.9
3.1
2.1
2.8
2.7
PEARSON
CORR.
COEF.
0.245
0.303
0.244
0.256
0.215
0.274
0.255
0.182
VARIANCE
LOWER
COMPARISON LIMIT
3.643
6.242
1.195
2.657
4.886
1.046
3.358
3.248
3.097
5.306
1.016
2.259
4.154
0.890
2.856
2.762
UPPER
LIMIT
4.284
7.342
1.406
3.125
5.747
1.230
3.949
3.820
SITE #6
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
264
268
263
263
273
265
263
AVERAGE
OBSERVED
VALUE
2.1
2.1
2.1
2.1
2.1
2.0
2.1
2.1
AVERAGE
LOWER
DIFFERENCE LIMIT
0.5
0.8
0.7
0.8
1.0
-0.8
0.7
0.8
0.
0.
0.
0.
0.
-1.
0.
0.
J
6
5
6
8
1
5
6
UPPER
LIMIT
0.7
1.0
0.9
1.0
1.1
-0.4
0,9
0.9
STANDARD
DEV. OF
RESIDUAL
1.7
1.6
1.7
1.9
1.3
3.0
1.5
1.5
LOWER
LIMIT
1.6
1.4
1 6
1 .7
1.2
2.8
1.4
1.4
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
UPPER
LIMIT
1.9
1.7
1.9
2.0
1.4
3.3
1.7
1.6
ROOT
MEAN SQ
ERROR
1.8
1.7
1.9
2.0
1.6
3.1
1.7
1.7
ABSOLUTE
RESIDUAL
1.4
1.4
1.5
1.6
1.3
2.2
1.3
1.3
PEARSON
CORR.
COEF.
0.527
0.416
0.239
0.379
0.551
0.275
0.492
0.511
VARIANCE
COMPARISON
0.502
0.939
1.107
0.581
1.183
0.224
0.814
0.832
LOWER
LIMIT
0.394
0.737
0.370
0.455
0.928
0.176
0.639
0.653
UPPER
LIMIT
0.640
1.197
1.407
0 .740
1.508
0.284
1.036
1.060
A-3
-------
TABLE A-2
HIGHEST OBSERVED AND PREDICTED CO CONCENTRATIONS (PPM)
EVENT BY EVENT (PAIRED IN TIME) USING MOBILE4.0
SITE#1
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
142
142
142
142
142
142
142
142
AVERAGE
OBSERVED
VALUE
5.8
5.8
5.8
5.8
5.8
5.8
5.8
5.8
AVERAGE
DIFFERENCE
4.0
4.8
2.2
3.3
4.6
0.6
4.0
3.2
LOWER
LIMIT
3.8
4.6
2.0
3.1
4.4
0.2
. 3.8
3.0
UPPER
LIMIT
4.2
4.9
2.4
3.5
4.8
1.0
4.2
3.5
STANDARD
DEV. OF
RESIDUAL
1.0
1.0
1.3
1.2
1.0
2.4
1.1
1.2"
LOWER
LIMIT
0.9
0.9
1.2
1.1
0.9-
2.2
1.0
1.1
UPPER
LIMIT
1.2
1.1
1.5
1 .4
1.1
2.7
1.2
1.4
SITE #2
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
143
143
143
143
143
143
143
143
AVERAGE
OBSERVED AVERAGE
VALUE DIFFERENCE
6.6
6.6
6.6
6.6
6.6
6.6
6.6
4.2
5.6
3.9
3.8
5.0
3.0
4.0
3.6
STANDARD
LOWER UPPER DEV. OF LOWER UPPER
LIMIT LIMIT RESIDUAL LIMIT LIMIT
3.9
5.4
3.7
3.6'
4.7
2.7
3.8
3.3
4.5
5.8 ,
4.2
4.1
5.2
3.3
4.3
3.8
1.6
1.2
1.4
1.6
1.4
2.0
' 1.5
1.5
1.4
1.1
1.3
1.5
1.2
1.8
1.4
1.4
i.a
1.4
1.6
1.9
1.5
2.2 .
1.7
1.7
SITE #3
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
66
66
66
66
66
66
66
66
AVERAGE
OBSERVED
VALUE
BTS
5.9
5.9
5.9
5.9
5.9
5.9
5.9
AVERAGE LOWER
DIFFERENCE LIMIT
T7T
3.7
0.3
0.5
1.6
0.6
0.6
1.5
0.3
3.4
-0.5
-0.7
0.9
-0.2
-0.0
0.8
UPPER
LIMIT
1.9
4.1
1.0
1.6
2.4
1.5
1.3
2.2
STANDARD
DEV. OF
RESIDUAL
3.3
1.4
3.0
4.7
3.0
3.4
2.7
2.8
LOWER
LIMIT
2.8
1.2
2.6
4.0
2.6
2.9
2.3
2.4
UPPER
LIMIT
4.0
1 .7
3 .7
5 .7
3.7
4 .1
3.2
3.4
A-4
-------
TABLE A-2 (continued)
HIGHEST OBSERVED AND PREDICTED CO CONCENTRATIONS (PPM)
EVENT BY EVENT (PAIRED IN TIME) USING MOBHJS4.0
SITE #4
MODEL
SPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
7~4
74
74
74
74
74
74
74
AVERAGE
OBSERVED
VALUE
o
5.7
5.7
5.7
• 5.7
5.7
5.7
5.7
AVERAGE
DIFFERENCE
2.0
3.8
1.2
1.5
3.1
0.6
1.1
3.1
LOWER
LIMIT
1.4
3.4
0.5
0.8
2.6
0.0
0.4
2.7
UPPER
LIMIT
2.6
4.1
1.3
2.2
3.5
1.2
1.8
3.5
STANDARD
DEV. OF
RESIDUAL
2.5
1.6
2.7
3.0
2.0
2.6
3.0
1.8
LOWER
LIMIT
2.2
1.3
2.4
2.6
1.7
2.2
2.6
1.6
UPPER
LIMIT
3.0
1.9
3.3
3.5
2.4
3.1
3.6
2.2
SITE #5
NUMBER AVERAGE
OF OBSERVED AVERAGE
MODEL EVENTS
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
75
75
75
75
75
75
75
VALUE DIFFERENCE
6.2
6.2
6.2
6.2
6.2
6.2
6.2
6.2
3.9
4.8.
1.9
3.3
4.6
1.5
4.0
3.7
LOWER
LIMIT
3.5
4.3
1.4
2.8
4.1
0.9
3.5
3.3
STANDARD
UPPER DEV. OF
LIMIT RESIDUAL
4.4
5.3
2.4
3.8
5.1
2.0
4.5
4.2
2.1
2.0
2.3
2.2
2.1
2.4
2.1
2.0
LOWER
LIMIT
1.8
1.7
1.9
1.9
1.8
2.0
1.8
1.7
UPPER
LIMIT
2.5
2.4
2.7
2.6
2.5
2.8
2.5
2.4
SITE #6
MODEL
SPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
NUMBER
OF
EVENTS
75
75
75
75
75
75
75
AVERAGE
OBSERVED
VALUE
4T5~~
4.0
4.0
4.0
4.0
4.0
4.0
4.0
AVERAGE
DIFFERENCE
-0..2
1.2
1.0
0.5
1.1
-2.4
0.6
0.5
LOWER
LIMIT
-0.7
0.8
0.6
-0.1
0.8
-3.1
0.2
0.2
UPPER
LIMIT
0.2
1.6
1.3
1.0
1.4
-1.7
1.0
0,8
STANDARD
DEV. OF
RESIDUAL
1.8
1.7
1.7
2.3
1.3
3.0
1.7
1.3
LOWER
LIMIT
1.6
1.5
1.4
2.0
1.2
2.6
1.5
1.1
UPPER
LIMIT
2.2
2.1
2.0
2.8
1.6
3.6
2.0
1.6
A-5
-------
TABLE A-3
HIGHEST OBSERVED AND PREDICTED CO CONCENTRATIONS (PPM)
PAIRED BY STATION USING MOBILE4.0
SITE#1
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
NUMBER
OF DATA
PAIRS
8
8
8
8
8
8
8
8
AVERAGE
OBSERVED
VALUE
8.4
8.4
8.4
8.4
8.4
8.4
8.4
8.4
AVERAGE LOWER UPPER
DIFFERENCE LIMIT LIMIT
4.9 3.6 £.3
6.0 4.9 7.2
2.6 0.9 4.4
3.8 2.4 5.3
5.7 4.5 6.9
-1.1 -4.1 1.9
5.1 3.7 6.5
4.0 2.8 5.1
STANDARD
DEV.OF
RESIDUAL
1.5
1.3
2.0
1.6
1.3
3.4
1.6
1.3
LOWER
LIMIT
1.0
0.9
1 3
1 i
0.9
2.2
1.0
0.3
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
UPPER
LIMIT
2.0
2.7
4.0
3.3
2.7
6.8
3.2
2.6
NUMBER
OF DATA
PAIRS
8
8
8
8
8
8
. 8
8
ROOT
MEAN SQ
ERROR
5.1
6.2
3.2
4.1
5.8
3.3
5.3
4.1
AVERAGE PEARSON
ABSOLUTE CORR. VARIANCE
RESIDUAL COEF. COMPARISON
'4.9 0.136 1.369
6.0 -0.229 11.953
2.6 -0.473 1.254
3.8 0.152 0.897
5.7 0.277 1.421
2.7 -0.246 0.181
5.1 -0.252 2.533
4.0 0.079 5.501
SITE #2
OBSERVED AVERAGE LOWER UPPER
VALUE
8.6
8.6
8.6
8.6
8.6
8.6
8.6
8.6
DIFFERENCE LIMIT LIMIT
5.4 3.3 7.5
7.2 5.6 8.9
5.3 3.1 7.5
5.4 3.2 ' 7.6
5.9 4.1 7.7
3.7 1.1 6.3
5.6 3.9 7.3
4.7 2.8 6.5
LOWER
LIMIT
0.274
2.393
0.251
0.180
0.284
0.036
0.507
1.101
STANDARD
DEV. OF
RESIDUAL
2.4
1.8
2.5
2.4
2.0
2.9
1.9
2.1
UPPER
LIMIT
6.838
59.707
6.263
4.481
7.097
0.905
12.651
27.477
LOWER
LIMIT
1.6
1 2
1 6
1 .6
1 3
1.9
1.3
1.4
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
UPPER
LIMIT
4.8
3.7
5.1
5.0
4.2
6.0
3.9
4.2
ROOT
MEAN SQ
ERROR
5.8
7.4
5.8
5.9
6.2
4.6
5.9
5.1
AVERAGE PEARSON
ABSOLUTE CORR. VARIANCE
RESIDUAL COEF. COMPARISON
5.4 0.202 1.070
7.2 0.297 15.341
5.3 0.331 0.649
5.4 0.268 0.801
5.9 0.167 2.796
3.7 0.309 0.428
5.6 0.292 2.526
4.7 0.155 2.506
LOWER
LIMIT
0.214
3.071
0.130
0.160
0.560
0.086
0.506
0.502
UPPER
LIMIT
5.344
76.629
3.243
4.003
13.964
2.137
12.618
12.519
A-6
-------
TABLE A-3 (continued)
HIGHEST OBSERVED AND PREDICTED CO CONCENTRATIONS (PPM)
PAIRED BY STATION USING MOBILE4.0
SITE #3
MODEL
3PAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
NUMBER AVERAGE
OF DATA OBSERVED AVERAGE LOWER
PAIRS VALUE DIFFERENCE LIMIT
8
8
3
3
3
8
8
8
8,
3,
8,
3.
8.
8.
8.
8,
, 5
.5
.5
,5
,5
,5
,5
,5
-1.0 -6.
5.1 3.
-1.0 -4 .
-4.3 -13.
-0.8 -5.
-0.3 -4.
-0.0 -2 .
0.5 -3.
b
4
5,
9
5
2
5
0
UPPER
LIMIT
4.5
6.7
2.5
5.3
4.0
3.5
2.5
3.9
STANDARD
DEV. OF
RESIDUAL
6.2
1.9
3.9
10.8
5.3
4.3
2.3
3.8
LOWER
LIMIT
4.1
1.2
2.6
7.1
3.5
2.9
1.8
2.5
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
UPPER
LIMIT
12.6
3.8
8.0
22.0
10.8
3.3
5.7
7.8
ROOT
MEAN SQ
ERROR
5.
5.
3.
11.
5.
4.
2.
3.
y
4
3
0
0
1
6
6
AVERAGE PEARSON
ABSOLUTE CORR.
RESIDUAL COEF .
4.6
5.1
2.8
7.3
3.9
3.1
2.2
3.2
0.389
-0.336
0.442
0.353
0.355
-0.034
0.082
0.288
VARIANCE
COMPARISON
0
0
0
0
0
0
0
0
.027
.764
.063
.009
.037
.067
.162
.072
LOWER
LIMIT
0.005
0.153
0.013
0.002
0.007
0.013
0.032
0.014
UPPER
LIMIT
0.136
3.816
0.316
0.046
0.185
0.335
0.810
0.361
SITE #4
MODEL
3PAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
NUMBER AVERAGE
OF DATA OBSERVED AVERAGE LOWER
PAIRS VALUE DIFFERENCE LIMIT
5
6
6
6
6
6
6
6
B.J
8.3
8.3
8.3
8.3
8.3
8.3
8.3
0
4
-0
0
2
1
-1
2
.9 -1
.3 2
.4 -3
.3 -4
.1 0
.0 -3
.1 -4
.0 -0
. b
.0
.8
.2
.7
.3
.2
.5
STANDARD
UPPER DEV. OF
LIMIT RESIDUAL
3,4
6.5
3.0
4.9
3.5
5.2
2.0
4.5
2.2
2.0
3.0
4.0
1.2
3.7
2.7
2.1
LOWER
LIMIT
1.4
1.2
1.8
2.5
0.8
2.3
1.7
1.3
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
UPPER
LIMIT
5.4
4.9
7.3
9.8
3.0
9.0
6.5
5.3
ROOT
MEAN SQ
ERROR
2 .
4.
2.
3.
2.
3.
2.
2.
2
6
7
6
4
5
7
8
ABSOLUTE
RESIDUAL
1.1
4.3
2.2
3.0
2.1
2.5
2.1
2.3
PEARSON
CORR.
COEF.
-0.239
-0.365
-0.494
0.198
0.155
-0.158
-0.060
-0.598
VARIANCE
COMPARISON
0
0
0
0
1
0
0
0
.321
.491
.175
.059
.209
.084
.166
.485
LOWER
LIMIT
0.045
0.069
0.024
0.008
0.169
0.012
0.023
0.068
UPPER
LIMIT
2
3
1
0
3
0
1
3
.297
.510
.249
.420
.637
.601
.189
.462
A-7
-------
TABLE A-3 (continued)
HIGHEST OBSERVED AND PREDICTED CO CONCENTRATIONS (PPM)
PAIRED BY STATION USING MOBILE4.0
SITE #5
MODEL
SPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
MODEL
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
NUMBER AVERAGE ~~
OF DATA OBSERVED AVERAGE LOWER UPPER
PAIRS VALUE DIFFERENCE LIMIT LIMIT
8
8
8
8
8
8
8
8
UPPER
LIMIT
6.6
5.2
7.5
7.6
6.5
6.8
7.0
7.5
NUMBER
OF DATA
PAIRS
4
4
4
4
4
4
4
4
UPPER
LIMIT
4.1
10.2
2.9
4.6
2.4
12.3
2.8
5.1
9.4
9.4
9.4
9.4
9.4
9.4
9.4
9.4
ROOT
MEAN SQ
ERROR
5.9
4.9
5.6
5.6
3.7
5.2
6.3
4.6 1
5.4 3
3.5 0
4.3 0
4.7 1
1.9 -1
4.1 1.
5.3 2,
AVERAGE
ABSOLUTE
RESIDUAL
4.5
5.4
3.5
4.3
4.7
2.8
4.1
5.3
- AVERAGE
OBSERVED AVERAGE
VALUE
4.3
4.3
4.3
4.3
4.3
4.3
4.3
4.3
ROOT
MEAN SQ
ERROR
1.1
2.9
0.8
1.2
1.3
7.8
0.7
1.3
PEARSON
CORR.
COEF.
-0.075
0.406
-0.416
0.045
0.165
0.124
0.048
-0.869
SITE #6
.7 7.4
.1 7.7
.2 6.8
.9 7.7
.8 7.5
.1 4.9
.1 7.2
.0 8.6
VARIANCE
COMPARISON
3.388
3.000
3.057
1.042
1.625
1.431
1.605
6.741
LOWER UPPER
DIFFERENCE LIMIT LIMIT
-0.5
0.3
-O..S
1.1
-7.3
-0.4
0.4
AVERAGE
ABSOLUTE
RESIDUAL
0.7
1.7
0.5
0.9
1.1
7.3
0.5
1.1
-2.
-6.
-1.
-2.
-0.
-13.
-1.
-2.
PEARSON
CORR.
COEF.
0.989
0.684
0.997
0.987
0.980
0.970
0.990
0.909
5 1.4
7 3.3
1 1.8
a 1.8
0 2.3
4 -1.2
7 1.0
0 2.9
VARIANCE
COMPARISON
0.536
0.515
1.946
0.501
1.407
0.211
0.662
2.572
STANDARD
DEV. OF LOWER
RESIDUAL LIMIT
3.2
2.6
3.7
3.8
3.2
3.4
3.4
3.7
LOWER
LIMIT
0.678
0.601
0.612
0.209
0.325
0.287
0.321
1.350
STANDARD
DEV. OF
RESIDUAL
1.1
2.7
0.8
1.2
0. 6
3.3
0.7
1.4
LOWER
LIMIT
0.035
0.033
0.126
0.032
0.091
0.014
0.043
0.167
2.1
1.7
2.4
2.5
2.1
2.2
2.3
2.4
UPPER
LIMIT
16.924
14.983
15.268
5.206
8.118
7.150
8.019
33.673
LOWER
LIMIT
0.6
1.6
0.4
0 7
0.4
1.9
0.4
0.8
UPPER
LIMIT
8.277 ' -
7.955
30.038
7.732
21.716
3.260
10.219
39.711
A-8
-------
TABLE A-4
25 HIGHEST OBSERVED AND PREDICTED CO CONCENTRATIONS (PPM)
UNPAIRED IN TIME OR LOCATION USING MOBILE4.0
SITE#1
AVERAGE AVERAGE
OBSERVED PREDICTED DIFFERENCE
MODEL VALUE VALUE OF AVERAGES
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
8.1
3.1
8.1
8.1
8.1
8.1
8.1
8.1
3
2
5
4
2
9
3
4
.1
.0
.3
.2
.3
.8
.1
.1
5
6
2
3
5
-1
4
3
.0
.1
.2
.3
.8
.3
.9
.9
LOWER
LIMIT
4.6
5.7
1.8
3.4
5.3
-2.3
4.6
3.5
UPPER VARIANCE
LIMIT COMPARISON
5.4
6.4
2.7
4.3
6.2
-1.2
5.3
4.3
1 .
6.
1.
1.
1.
0.
4.
3.
946
654
684
674
573
616
778
249
LOWER
LIMIT
0.857
2.931
0.742
0.738
0.693
0.271
2.105
1.431
UPPER
LIMIT
4.412
15.088
3.819
3.797
3.567
1.397
10.834
7.367
SITE #2
AVERAGE AVERAGE
OBSERVED PREDICTED DIFFERENCE
MODEL VALUE VALUE OF AVERAGES
EPAINT
FHWAINT
C ALINE 4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
8.6
8.6
3.6
3.6
8.6
3.6
3.6
3.6
4
1
4
4
3
6
4
4
.5
.6-
.5
.9
.0
.7
.1
.6
4.
6.
4.
3.
5.
1.
4.
4.
0
9
1
7
5
9
5
0
LOWER
LIMIT
3.4
. 6.5
3.5
3.1
5.0
1.1
4.0
3.5
UPPER VARIANCE LOWER
LIMIT COMPARISON LIMIT
4.6
7.4
4.7
4.3
6.0
2.7
5.0
4.5
1
15
1
1
4
0
5
5
. 662
.989
.788
.285
.875
.418
.339
.071
0.732
7.044
0.787
0.566
2.148
0.184
2.352
2.234
UPPER
LIMIT
3.769
36.257
4.054
2.913
11.055
0.949
12.107
11.499
SITE #3
OBSERVED PREDICTED DIFFERENCE
MODEL VALUE VALUE OF AVERAGES
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
7.8
7.8
7.8
7.8
7,8
7.8
7.8
7.8
8.9
3.5
9.5
11.5
7.8
9.0
8.7
7.8
-1.1
4.2
-1.8
-3.8
-0.1
-1.3
-0.9
0.0
LOWER
LIMIT
-2.6
3.7
-2.8
-6.2
-1.4
-2.3
-1.7
-1.0
UPPER VARIANCE
LIMIT COMPARISON
0.4
4.7
-0.8
-1.3
1.3
-0.2
-0.2
1.1
0.084
2.340
0.216
0.031
0.102
0.199
0.499
0.191
LOWER
LIMIT
0.037
1.031
0.095
0.013
0.045
0.088
0.220
0.084
UPPER
LIMIT
0.190
5.307
0.490
0.069
0.231
0.452
1.132
0.434
A-9
-------
TABLE A-4 (continued)
25 HIGHEST OBSERVED AND PREDICTED CO CONCENTRATIONS (PPM)
UNPAIRED IN TIME OR LOCATION USING MOBILES
SITE #4
MODEL
PHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
MODEL
EPAINT
PHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
OBSERVED
VALUE
7.4
7.4
7.4
7.4
7.4
7.4
7.4
OBSERVED
VALUE
8.8
8.8
8.8
8.8
8.8
8.8
8.8
8.8
AVERAGE
PREDICTED
VALUE
7.1
3.6
8.3
8.1
5.4
8.2
8.3
4.6
AVERAGE
PREDICTED
VALUE
3.9
3.0
6.4
4.6
3.5
7.0
4.0
3.9
DIFFERENCE
OF AVERAGES
0.3
3.8
-0.9
-0.7
2.0
-0.8
-0.9
2.8
DIFFERENCE
OF AVERAGES
4.8
5.8
2.4
4.1
5.3
1.8
4.8
4.9
LOWER
LIMIT
-0.3
3.3
-1.5
-1.5
1.5
-1.6
-1.7
2.2
SITE #5
LOWER
LIMIT
3.9
4.8
1.5
3.1
4.3
o.a
3.7
4.0
UPPER
LIMIT
0.8
4.3
-0.3
0.2
2.5
-0.0
-0.1
3.4
UPPER
LIMIT
5.8
6.7
3.3
5'.2
6.4
2.8
5.8
5.7
VARIANCE
COMPARISON
0.694
0.840
0.510
0.196
0.938
0.245
0.210
0.411
VARIANCE
COMPARISON
3.085
3.446
6.936
2.150
1.671
2.081
1.905
14.677
LOWER
LIMIT
0.306
0.370
0.225
0.086
0.413
0.108
0.093
0.181
LOWER
LIMIT
1.359
1.518
3.056
0.947
0.736
0.917
0.839
6.466
UPPER
LIMIT
1.575
1.904
1.157
'0.443
2.126
0.555
0.477
0.932
UPPER
LIMIT
6.997
7.815
15.728
4.876
3.789
4.720
4.319
33.282
SITE #6
OBSERVED PREDICTED DIFFERENCE LOWER
MODEL VALUE VALUE OF AVERAGES LIMIT
EPAINT
FHWAINT
CALINE4
CAL3QHC
VOL9MOB4
TEXIN2
GIM
IMM4
4.9
4.9
4.9
4.9
4.9
4.9
4.9
4.9
6.4
4.5
4.4
6.2
4.1
9.3
4.9
4.9
-1.4
0.5
0.5
-1.2
0.9
-4.4
0.0
0.1
-2.1
-0.4
-0.1
-2.0
0.3
-5.6
-0.7
-0.5
UPPER VARIANCE LOWER
LIMIT COMPARISON LIMIT
-0.7
1.3
1.1
-0.4
1.4
-3.1
0.8
0.6
0.810
0.421
1.311
0.515
2.234
0.166
0.679
2.319
0.357
0.186
0.577
0.227
0.984
0.073
0.299
1.022
UPPER
LIMIT
1.836
0.956
2.972
1. 167
5.066
0.376
1.540
5.258
A-10
-------
g
>-H
CO
D
CO
z
o
p
S P
H ^
Z £™
S 2^
072
w
Si
8s
;: a
Q g
rTl *^*^
C CO
U D
>— i .
Q >«
W O
S3
ffi O
G |*V]
>-* o
^i
a S
"? Ss
« go £
W H
£g
2S
o <
^^
PU ^
O ^
Z S
O
CO
}-H
^
£
^
0
u
<
«s.
U 0.
?fi
s s
iE
£a
sa
IxB
'SB S;
u
Sse
CJ t3 ft
•SSH
gsa
zui
•?s
u a.
?fi
5 a
i §.
ia
S^
S a
S a!
Sfi a
S
SSE
e""
5gfE
« O CU
3
sss-
01
Ul
y
N^
s«
rfi
gB
t
a
s
a«AW\0^^
•Vtn>-l<*1C4<*)(S(MfON
r»WNtoaw>«H«nt0o
Mp^^H.^^ltHtHiH*^^
\0F>^>mm4«4«aaa
nr'liH^c^cit-icsicM^
\o*vftt-><&*DOf*t*in
r*«v(MMWtr>
n4i
r^m«y«a«9(Mtn<*«*N<*i^>-^>(MUl
-«i-«csi
Men-Hf-tcnrti-ttnntn
aoouir»«HQ9a\aoA
«««««„««<««
c9«ocBa>a>««>matA
04 (N rt CM fH '&*rtrtme^
$
1
ft— i
CO
Jg
da
IE
5 a
g§
Si:
£E
u 0.
2SE
S5E
«oa
•gSE
u > o.
3SE
su a
^E
5s
IE
Ss
? 8.
&i a*
£E
Sa
u E
M Q.
] sa.
o&
en O.
o a
5) J3 3.
rt
^
^
U
01
2*0.
vj S
M e
5-S
ss
« w
Q -a
5J °*
SS1
§•"
s.
u?
xo
X
5
i
N«si«n-^*-«n*oi^i-«^
r>«C44nr«i-t>Afn«»«*AO1Oi*m
rt^r>«0wm
•*M»W«tn(MC4^>iAU>lA
i*)OOA^WC4(nO
i-tMW*-t^t*Hi-ti(n
**t^W(nwu>w(MAc4tnru«r^u>o
(*mM^NNd«4\OC4(4C4«4CSI(MMr»vir»«(»>
n^Otnt
?xg
fa
52i
|5a
flS B
n > a*
CO
flOH
SQ|
-J g
gg
5 g.
si
£a
s t
3 a.
M E
Hft
•35 a
«a
«s4
SSE
•a o o.
3
|*5 S,
M
u
o
WJ
X
5-S.
sa
N Dl
S o
a11
3?
§'
lh
9
= m
>*
>»
Q
§
v>^>N(nui^«n««^y>
AMemoe^AVocs
nwr-mr-mAWUi
^^<-««H(Moir»«pm«4(nio
«>i~ic>*-4U9VD^wi^i«titon
r»fH4«*>Mm
war-fK^otr^atcBp*
tnwlMr^C^^^Ot-A
or-iA
co0tWAu>r-Mt-*r-*n*o^«-iwu>a>vio
*n*N*o
0tt*-r»ut*a>M>w
^t *H C4 <»cocQ
m = stability class al
en
ex
6
CO
e*f
o S
«" >3
— <
So
«j
s^
s o
lg
"° s
*1
"?
ss
M
;-a
w II
-s
2 £3
o H
S aa
a ^
direction
; TEX =
•f s
* s
! 2
£o
es ^
-, II
S^
S o\
9>
^c7
i~|
l|
s o
S II
"cy
9 ^
: temperature o\
= Observed; (
iM
s w
'i S
Note: TAVG
monitor 1 , 2,
A-ll
-------
g
)— 4
D
CO
0
<
§5
Pl^
O ro
P
C")
§i
g|
^ lo
•g p ttj
s O ^
O pH /-\ •"*•
^ £« *
•9 S > S
L t™1 «y t"*
^r-o^*«,««
•*r4NOv**4^m.-4<<4
-i!»o«ri"io
tn*DFl*tcM.n«*atnr>>
r4^N*-*r4OAr4O|U>
«,or--«.c.,^.
rt F^
o.c-o»r,r-^.n
»<.««rt«ona>«
~.«a«.
or-a>a«n in in in va n n u>
NACMWWOmOWA
^w.^o^r,-,-,™
m<-«c4
..,, ,„„»„„,
— -
^(*w»c**He*c**nmc*
mvu>r»«l«v
«.no»-«o>mP.-
<-***OWli-**OCMO^W
ss^ssSoisss
CM CM
* -« rt 3*
^ bl Q.
ot.a
«SE
quo,
aS S.
3 a
CU
IE
SE
n a.
ii5E
«_i
»z E
u
o ct e
§..s
S§l
S
^ J3
ga
3 ^
SB
II
11
> tu
14 td
303
U
O
A- 12
f****0fMr*4ncnc*f«witn^iu>^^>
«*-4Nmn(MC4C4NC
« «o
«o ^ « en N * in
m^«a<»«»f*H»«nMMnn
N**«n«^«.*-i«
*itnmn^uin
^ Q*
ssa
•« O Q.
I. a
o ja 3,
a: o a
o
s s
3S1
a "a
§„
^0.
££
3-
n«n
oiimrrm
P- *D 1* *0 «• Ain«\r-mc«ot
Kiawuvuiinuiinui
2 3 "2 "3°'::°' 2
•"•-•"» ""•"">
« = wiiu
JHC; Y9M = VQL9MOE
S <°.
la
a it
wCC
SC)
II
&.£!
s o
ai
"5
ll
-------
§
eo
2
55
CO
o»o» • !* ~r
V ^V-K\
v\ --Vsx
A \V \\
\ \ ^v x
\ \ \ \
\ \ ^ \
\ \ ^. \
A O» • «-!«•
,o en
O a>
s . e
§••8
S
A-13
-------
(D
CO
Or
as;
CO-
CO
9101 >*4«
mo • oo
- in0» *mr
A A " r*t*
««• * -*m
£MS
a.-a
.2 -
§3
§ ^
853
"8 «
•8
c "«
§ 3
3 T3
*
II
cs
1
U)
S
A-14
-------
3
2
55
2
S
3 ctf
CO
S
\.
\
\
\ 1
S
A
2 ,
§
A-15
-------
to
CO
BbKUOMUHMQ
X
rfi
1.1 \
"SS
r°
'•o
(U
"S
S §
CD
s=:
03
\
O.KUOUUMMQ
;
'I A
IT
T,
'•!i
T^T
S 35
"O s
«
o 2
'•3 5
2 '« —
a, .2 ys
«o 2 x
S O J^
& (33
.
SO 3 ~~
3 c3
2 "3 S
s13 > C
E-1
-------
1
s
Bi
0»
a
o
§
o .5
2 =
0 -ri
"S, «^
33 2
SI
.W3 O
So
O-CBCUOMU H WQ
> 0. o. B
A-17
-------
^^r-i_ in
I
I
CM
CO
I
§
s
1
£
cu
2
o
u>
m
o
1
11
2"!
«
o .5
03
=2 =s
"3.13
CO 4J
<
2
3
SO
O.rt UQHO 6-"UQ
A-18
-------
j£
d
I
CO
O-OSMOWO^BIQ OOBOWBPOS-JUHOS: I a. O. a
•* a
0.
o.
1
o
n
01
o g
"I
as
e en
8=*=
§.3
O CO
I?
13
& «
Q
i2 S
° T3
a, o
« 2
11
CO O
A-19
-------
I
g
04
- in I
ui
01
o
O.C!MQHOtlMQ O O 3 O U B &4 Pi rt ^ M O 2 I Ot Ck X
•3
g
"I
o
o -~
CJ CO
•8
o
2
••8 o
2 '33
3. .
3
-------
s
8
0,
§
H
ej
53
u
8
0}
0}
o
•3
o
1
.s
§
-4»»
£
a
o -S
O CO
o g
"-3 o
1.1
„ =
3
-------
I
CD
O
•a
o
03
I
4«ft
£ .
C3 ^O
It
85
09
3
-------
APPENDIX B
RESIDUAL PLOTS USING MOBELE4.1 EMISSIONS
-------
-------
SITE 1
tu
1 010!
0.10
0.01
fj. A-
T
I,
£ • a
y p y
3 12 16 20
HOUR OF THE DRY
10.00
1.00
x1 0.10
uj
0.01
90 180 270
HIND DIRECTION
10.00
1.00
x 0.10
UJ
0.01
4 S
U (HPHJ
10.00
10
25 50 75
TEMPERHTURE
100
10.00
1.00
x 0.10
LU
0.01
rA o
10.00
§ 1.00
>x
X 0.10
IH
0.01
*
PG OflSS
0 2000 4000 6000 S000 10000 12000 H00S
TRRFFIC VOLUME
Figure B-l. The residual or ratio of the predicted to observed concentration using the
TEXIN2 model with MOBUJE4.1 emissions at Site #1 plotted versus the hour
of the day, wind direction, wind speed (u), ambient temperature, Pasquill-Gifford
(PG) stability class, and traffic volume. Significant points on each box plot
represent the 2nd, 16th, 50th, 84th, and 98th percentiles. The number of
observations used in each box are also labelled near the bottom as "N = #." The
dashed lines represent the factor of two lines.
B-l
-------
SITE 1
i flfl
0.10
n.cn
._S JL A i.__
v & y V
1
•
JF ^i> ^ ^>
A
v
^
en
§
^
en
g
in. aa
i an
0.10
a. HI
I A fi
T 9
^ ^ ^
8 12 16 20 21
HOUR OF THE DflY
90 ' 180 270
MIND DIRECTION
1.00
0.10
0.01
...........id.......
I....Q—g—-g
8 10
U (HPHJ
10.00
en
CO
0 1.00
0.10
0.01
.J5...A....S....JL...A
25 50 75
TEMPERHTURE
10.00
1.00
0.10
0.01
PG CLflSS
10.00
1.00
0.10
0.01
0 2000 4000 6000 8000 10000 12000
TRRFFIC VOLUME
Figure B-2. Same as Figure B-l except for the CAL3QHC model.
B-2
-------
SITE 1
§1 nn
1.00
5 0.10
0.01
u H. tjjl iy A...
I
10.00
o 1.00
£
z
~ 0.10
a ai
1 T
^ #* Jk J> ^
0 * 3 12 IS 20 24 0 g0 i80 270 3£
jg
HOUR OF THE DRY HIND OIRECnoN
10.00
§ i 00
5 0.10
LJ
0.01
.. a...^9..|.^..._....
,** & /$> ,<$> & 3> Ns
•v -y ^ ^ -^ ^ •&
?> ^fr ^V
0 2 4 6 8 10 0 25 50 75 100
U (MPH) TEMPERHTURE
10.00
§ 1.00
55
~ 0.10
s
0.01
.........
fl—f—f-— cr-Q-
tf
•^ ^ ^" ^' -^
in nn
in
§ 1.00
>.
<»•
^
2 0.10
QJ
•••-••.•«"•.••-—««•-«.•..»....•..„...•„„. .
g..g..5..J.jj..l.
^^^^s/^^
* 30 2000 4000 6000 3000 10000 12000 140
00
PG CLASS TBOPPrr vni iiuc
Figure B-3. Same as Figure B-l except for the CALINE4 model.
B-3
-------
SITE 1
i 0.10
0.01
**
0 4 8 12 16 20 24
HOUR OF THE DRY
10.00
in
g 1.00
0.10
0.01
1^
90 180 270 360
HIND DIRECTION
10.00
1.00
0.10
0.01
*..— W.««rg....A
A r
248
U (HPH)
8 10
10.00
in
g 1.00
v.
i 0.10
0.01
«•.•••••«•••••«..••••.I-i««»4^VB«»Jfr-j-"«4g-j*»"^^ ••
25 50 . 75
TEMPERflTURE
100
10.00
1.20
0.10
0.01
1
PG CLflSS
10.00
i 1.00
s
i 0.10
0.01
2000 4000 6000 8000 10000 12000 14000
TRRFFIC VOLUME
Figure B-4. Same as Figure B-l except for the IMM model.
B-4
-------
SITE 1
10.00
1.00
a 0.10
0.01
10.00
§ 1.00
3 0.10
0.01
10.00
g 1.00
>
3 0.10
0.01
3 12 16 20
HOUR OF THE DRY
H"-?-^""g""5""g--J
0 2 4 6 8 10
U (MPH)
-••••»-.••....
S 5 5 e s
f
4
PG OflSS
10.00
§ 1.00
3 0.10
0.01
10.00
§ 1.00
3 0.10
0.01
10.00
g 1.00
^
3 0.10
0.01
,..T-
90 180 270
HIND DIRECTION
3S0
.—.................._... -a.-—A—--I-....—.............
5 * f a 9
0 25 50 75 100
TEMPERATURE
0 2000 4000 8000 8000 10000 12000 14000
TRRFFIC VOLUME
Figure B-5. Same as Figure B-l except for the GIM model.
B-5
-------
SITE 2
S"
10.83
1.03
0.10
— — —
y—n.—A—g—u.~.
^
8 12 18
HOUR OF THE OHY
10.00
8 1.00
3 0.10
*-
0.01
10.00
§ 1.00
S
| a-10
0.01
A
U (MPH3
4
PG CLflSS
20 24
10
10.00
1.00
0.10
0.01
10.00
en
S 1.00
I
x 0.10
0.01
10.00
in
S 1.00
0.10
A 1 "
...|
•f-*—
90 180 270
WIND DIRECTION
360
T"T'"i T"'
H H D J5
25 50 7S
TEHPERflTURE
100
0.01
fj n M
0 2000 4000 6000 8000 10000 12000 H00EI
TRRFFIC VOLUME
Figure B-6. Same as Figure B-l except for the TEXIN2 model at Site #2.
B-6
-------
SITE 2
in
§ 1.00
V.
g
0.01
10.00
° 1.00
£
a
3 0.10
0.01
0
10,00 r
tri
1.00 •
X
3 0.10
cz
u
0.01 L
0
9 t fl J a
•>** <^ -$?* ^ *
3 * 3 12 IS 20 2
HOUR OF THE DBY
* $ ? 5~* '
**•>•*«>*
^ ^ ^x ^x -^ ^%
2 4 S 8 i(
U (MPHJ
'""9" * "
by^ O?^ fcP^ %
— ^ — £ — ^ ^ ^s
4 o
-} 10.00
l.iJH
§ 0.10
» 3'01
CO
S i 00
5 0.10
5
> a
(n
1.00
zs
a
3 0.10
(Z
u
.01 u
.... .
~TTJHf-H-ir-
^sS ^ ^ J* J> , 1 1 ^ —
•&* ^j^"^ J$
25 50 75 100
TEMPERflTURE
&-A.i
M T •
^^^^
PG CUflSS
0 2000 4000 6000 8000 10000 12000
TRBFFIC VOLUME
Figure B-7. Same as Figure B-l except for the CAL3QHC model at Site #2.
B-7
-------
SITE 2
I
3
u
10.00
1.00
19
rt
.
(9
0.01
8 12 16
HOUR OF THE DRY
20 24
1.00
5 0.10
a.t
30 -180 270
NINO DIRECTION
360
10.00
1.00
s 0.10
0.01
16
U IHPHJ
10
10.00
5 0.10
a. 01
•••••••••••••••••••••••Mr ••••« -.•••J»«t««»^3»« ••***••
f fl g
25 50 75
TEHPERflTURE
100
10.00
1.00
= 0.10
a!
u
0.01
-H5rt~"t
^ ,&? n
;jr°r
10.£
S 1.00
w
5 0.10
a.i
PG CLHSS
2000 4000 6000 8000 10000 12000 14000
TRRFFIC VOLUME
Figure B-8. Same as Figure B-l except for the CALINE4 model at Site #2.
B-S
-------
SITE 2
§ 1 00
i 0.10
0.01
!
10.00
en
§ 1 00
§
1 0.10
0.01
0
10.00
in
g 1.00
i 0.10
0.01
a
^ r3> 4>' rgt <<)
•& •& •& •& •&
3 4 8 12 16 20 2
HOUR OF THE OflY
^ jfy ^ # ^p ^
.2 4 S 8 11
U (MPH)
sl*,l I
10.00
tn
• S i«00
I 0.10
3 !
in nn
en
§ 1.00
i
i 0.10
am
3 2
s
i 0- 10
0.01
_rr^^_
5 30 180 270 36
MIND DIRECTION
"*"™*"*"*""*""""'^"""««*"nSJ«»»«^3«»«fft«««««*»JTiMB-,
•19
25 50 75 10E
TEMPERflTURE
a 9 98
PG CLflSS
0 2000 1000 6000 3000 10000 12000 11000
TRRFFIC VOLUME
Figure B-9. Same as Figure B-l except for the IMM model at Site #2.
B-9
-------
SITE 2
10.00
in
g 1.00
3 0.10
0.01
e
10.00
in
g 1.00
3 0.10
0.01
£
10.00
4
1.23
o 0.10
0.01
i j. 4 _,
g- f fl ? fl
P} t& «^ IX* ^
*^* -j& "\y? -*y ~^f
4 8 12 16 20 2<
HOUR OF THE DRY
JL _ j. •
*}}}*}*
i 2 4 S 8 1
U 1HPH)
• _ Ji.
g 9 * 9 T
a 4
PG OflSS
iv.wa
§t cm
3 0.10
ant
i e
m ctn
S. »M
O
S 0.10
0 01
0 S
in aa
o 1 00
0 "
3 0.10
a fill
3
j«, A __^
^v ^b ^^ &) ^JV ^>
•^f *s» <^^ -^f •$& •$
90 • 180 270 36
HIND DIRECTION
•O -A- 1Q T*I A««
S f#f*
J 2S 50 75 10
TEMPERflTURE
_^ JL JL
fw
3 2000 4000 6000 8000 10000 12000 140
TRflFFIC VOLUME
Figure B-10. Same as Figure B-l except for the GIM model at Site #2.
B-10
-------
SITE 5
i 1.00
3 0.10
ft!
0.01
S
10.00
i 1.00
55
0.01
0
10.00
cri
§ 1.00
i
x 0.10
ft!
0.01
H i*r a
^ H tl
' t 3 12 16 20 2
HOUR OF THE DHY
........ j „ _„.
^~r~~~~
2 i s a u
U tMPHJ
j • •
10.00
i 0.10
ft!
0.01
3 , (
10.00
en
§ 1.00
§ 0.10
0.01
9 0
id ctcv
en
§ 1.00
x 0.10
ft!
0.01
' 30 180 270 35
MINO DIRECTION
V -
25 50 75 10E
TEHPERFnrURE
n _ ir~jL x ~~'—~
.^. J.j..g...u..6
0 2000
PG CUBS
5000 3000 10000 12000 H000
TRRFFIC VOLUME
Figure B-ll. Same as Figure B-l except for the TEXIN2 model at Site #5.
B-ll
-------
SITE 5
.
o i aa
U
| 0.10
g
0 01
p A ..™-fl..™.,a_ m
U V g y
/5* «,*& S& x >
•^ "& ^ •^•'
tn
i ari
« 0.12
u
a ai
r i
fA
f
v/V* X y-S^ X*^ /I/
V ^ ^ "^ "S"'
0 4 8 12 IS 20 23 a 90 180 270 3S
HOUR OF THE Qflr WIND DIRECTION
en
§ a. 10
£
3SH
1 1 ,
V I | 5
% ^> *. o, A
^? ^f?1' ^ ^? ^?
1CJ. 4JVJ
tri
1.00
. § 0.10
g
a a*
...U...fl........««.....
If f
* # /o
^? -^r ^r'
0 2 4 6 a 10 3 25 53 75 10
U IMFHJ TEMPERflTURE
i n tin 1 ™ ""
en
S i.aa
^. 0. 10
<£
U
fl.fll
, „. . &.. A A.
* tf § 1
r[J> A «&
^J* ^ ^?
4U* UU
1.20
i
s
-------
SITE 5
S 1.20
2 3.10
0.21
t
10.00
in
a 1.20
- a. 10
LJ
0.21
e
10.00
S 1.20
§
S 0.10
0.21
• JT _ . -
j$> j$> j$> ^>
! 1 3 12 16 20 2
HOUR OF THE DRY
n S • -, "
2.4 S 3 11
U (MPHJ
JL •
^ ** J>
m
•N.
%
ca
CRL1NE4/ (
3
tn
\
UJ
10.20
0.10
(
0.10
d ni
E
in an
3.10
3.31
,...«.«««,_i™ —
4 ""~ifl~~ I — a i
5 90 180 270 3S
WIND DIRECTION
~ ~ ~ ™~— -jj-™~— ™~— — -
^...^ „
25 50 75 101
TEMPERHTURE
- -..-.^...........^j^a
7!Zr
PG OflSS
0 2000 ^000 8000 8000 10002 12000 14000
VOLUME
Figure B-13. Same as Figure B-l except for the CALINE4 model at Site #5.
B-13
-------
SITE 5
10.00
* 0.10
0.J
JL Jj JL __ A
0 4 8 12 IS 20 24
HOUR OF THE OflY
10.30
1.20
1 3.10
a. 01
A
90 180 270
HIND DIRECTION
360
10.00
1.00
0.10
0.01
.£„..£
|= 1
TO
0248
U 1MPH)
3 10
1.00
0o 12
8.E
25 50 75
TEHPERflTURE
100
10.00
1
2 8.10
0.01
1
PG cuss
3.00
s.
1 0.10
a.:
0 2000 1000 6000 3000 10000 12000 H00I3
TRRFFIC VOLUME
Figure B-14. Same as Figure B-l except for the IMM model at Site #5.
B-14
-------
SITE 5
10.00
1.213
a 0.12
0.01
10.00
1.00
3.10
0.01
10.00
S 1.00
0.10
0.01
_•_ 1
3 4 3 12 IS 20 2
HOUR OF THE DRY
1 | s "
5 2 4 3 3 11
U tHPHJ
A . _ I
f 9 g
in.uu
S i 0H
•x.
0 0.10
ant
i t
tfl an
S t 00
3 0.10
3 0
in ctn
§1 an
5 0.10
3.21
n
§H A «
— ~ — •» — n — •* — Q~
v y T
1 90 180 270 35
MIND DIRECTION
1
"? y^1 /£•
25 50 75 10
TEHPERBTORE
I
t n JL 3. JL
I* 5*Lw
PG OflSS
0 2000 3000 5000 3000 10000 12000 14000
TRBFFIC VOLUME
Figure B-15. Same as Figure B-l except for the GIM model at Site #5.
B-15
-------
-------
TECHNICAL REPORT DATA ~
(Pleas* read Instructions on the reverse before completing)
EPA-454/R-92-Q04
2.
3. RECIPIENT'S ACCESSION NO.
..TITLE AND SUBTITLE
Evaluation of CO Intersection Modeling Techniques Using
a New York City Database
5. REPORT DATE
August 1992
6. PERFORMING ORGANIZATION CODE
B.C. DiCristofaro, D.G. Strimaitis, and R.C. Mentzer
8. PERFORMING ORGANIZATION REPORT NO.
DON NAME AND ADDRESS
10. PROGRAM ELEMENT NO.
Sigma Research Corporation
Concord, MA 01742
li. CONTRACT/GRANT NO.
12. SPONSORING AGENCY NAME AND ADDRESS
Office of Air Quality Planning and Standards
U.S..Environmental Protection Agency
Research Triangle Park, NC 27711
13. TYPE OF REPORT AND PERIOD COVERED
14. SPONSORING AGENCY CODE
68A
-------
-------
|