United States Office of
Environmental Protection Research and Development
Agency Washington, DC 20460
EPA-600/R-98-097
August 1998
m EPA A GIS-Based
Modal Model
Of Automobile
Exhaust Emissions
Prepared for National Risk Management Research Laboratory
Prepared by Air Pollution Prevention and Control Division
-------
FOREWORD
The U. S. Environmental Protection Agency is charged by Congress with pro-
.tecting the Nation's land, air, and water resources. Under a mandate of national
environmental laws, the Agency strives to formulate and implement actions lead-
ing to a compatible balance .between human activities and the ability of natural
systems to support and nurture life. To meet this mandate, EPA's research
program is providing data and technical support for solving environmental.pro-
blems today and building a science knowledge base necessary to manage our eco-
logical resources wisely, understand how pollutants affect our health, and pre-
vent or reduce environmental risks in the future.
The National Risk Management Research Laboratory is the Agency's center for
investigation of technological and management approaches for reducing risks
from threats to human health and the environment. The focus of the Laboratory's
research program is on methods for the prevention and control of pollution to air,
land, water, and subsurface resources; protection of water quality in public water
systems; remediation of contaminated sites and groundwater; and prevention and
control of indoor air pollution. The goal of this research effort is to catalyze
development and implementation of innovative, cost-effective environmental
technologies; develop scientific and engineering information needed by EPA to
support regulatory and policy decisions; and provide technical support and infor-
mation transfer to ensure effective implementation of environmental regulations
and strategies.
This publication has been produced as part of the Laboratory's strategic long-
term research plan. It is published and made available by EPA's Office of Re-
search and Development to assist the user community and to link researchers
with their clients.
E. Timothy Oppelt, Director
National Risk Management Research Laboratory
EPA REVIEW NOTICE
This report has been peer and administratively reviewed by the U.S. Environmental
Protection Agency, and approved for publication. Mention of trade names or
commercial products does not constitute endorsement or recommendation for use.
This document is available to the public through the National Technical Information
Service, Springfield, Virginia 22161.
-------
EPA-600/R-98-097
August 1998
A GIS-Based Modal Model of Automobile
Exhaust Emissions
FINAL REPORT
Prepared by:
William H. Bachman
School of Civil and Environmental Engineering
Center for Geographic Information Systems
Georgia Institute of Technology
Atlanta, Georgia 30332
EPA Cooperative Agreement CR823020
Project Officer: Carl T. Ripberger
U.S. Environmental Protection Agency
Air Pollution Prevention and Control Division
Research Triangle Park, North Carolina 27711
Prepared for:
U.S. Environmental Protection Agency
Office of Research and Development
Washington, DC 20460
-------
ABSTRACT
Suburban sprawl, population growth, and auto-dependency have, along with
other factors, been linked to air pollution problems in U.S metropolitan areas.
Addressing these problems becomes difficult when trying to accommodate the needs
of a growing population and economy while simultaneously lowering or maintaining
levels of ambient pollutants. • Growing urban areas must, therefore, continually
develop creative strategies to curb increased pollutant production.
This report presents progress towards the development of a computer tool
called MEASURE, the Mobile Emission Assessment System for Urban and Regional
Evaluation. The tool works towards a goal of providing researchers and planners with
a means for assessing new mobile emission mitigation strategies. The model is based
in a geographic information system (GIS) and uses modal emission rates, varying
emissions according to vehicle technologies and modal operation (acceleration,
deceleration, cruise, and idle). Estimates of spatially resolved fleet composition and
activity are combined with situation-specific emission rates to predict engine start and
running exhaust emissions. The estimates are provided at user-defined spatial scales.
A demonstration of model operation is provided using a 100 square kilometer study
area located in Atlanta, Georgia. Future mobile emissions modeling research needs
are developed from an analysis of the sources of model error.
-------
TABLE OF CONTENTS
ABSTRACT ...ii
TABLE OF FIGURES • .»» vii
GLOSSARY AND ACRONYMS ix
1. INTRODUCTION 1
1.1. SUMMARY OF CONTRIBUTIONS TO RESEARCH 5
1.2. REPORT ORGANIZATION 6
2. BACKGROUND 7
2.1. AUTOMOBILE EXHAUST EMISSIONS 7
2.7.7. Exhaust Emission Pollutants 7
2.1.2. The Mechanics of Exhaust Emissions 8
2.1.3. Modal Emissions 9
2.1.3.1. Emissions in Start Mode 10
2.1.3.2. Emissions in Hot-stabilized Mode 12
2.1.3.3. Off-Cycle Exhaust Emissions 13
2.2. AUTOMOBILE EXHAUST EMISSION RATE PREDICTION 13
2.2.7. A Speed Correction Factor Approach 14
2.2.2. A Physical Approach 75
2.2.5. A Statistical Approach 76
2.2.4. Emission Rate Modeling Summary 77
2.3 VEHICLE ACTIVITY MODELING 18
2.3.7. Urban Transportation Planning System (UTPS) 18
2.3.2. Simulation Models (TRANSIMS) 79
2:4. GEOGRAPHIC INFORMATION SYSTEMS 20
2.4.1. CIS in the Transportation /Air Quality Agencies 27
2.4.2. Applications of CIS in Mobile Emission Modeling 21
2.4.2.1. Emission Inventories 21
2.4.2.2. GIS for Transportation Planning and Air Quality Analysis 22
2.4.2.3. Microscale Analysis 22.
2.4.2.4. Influencing Decision-makers 22
2.4.3. Spatial Data Issues 23
2.4.3.1. Positional Accuracy 23
2.4.3.2. Data Resolution 23
2.4.3.3. Data Content Accuracy 24
3. MODEL CONCEPTUAL DESIGN 25
3.1. MODEL DESIGN PARAMETERS 26
3.2. USER REQUIREMENTS 29
3.3. THE SPATIAL DATA MODEL 31
in
-------
3.4. MODEL APPROACH , 33
3.4.1. Spatial Environment r 35
3.4.1.1. Zonal Data 35
3.4.1.2. Lineal Data • 36
3.4.1.3. Conflation 36
3.4.2. Fleet Characteristics • 37
3.4.2.1. Vehicle Geocoding 38
3.4.2.2. VIN Decoding 38
3.4.2.3. High and Normal Emitters 39
3.4.2.4. Technology Groups : 40
3.4.3. Vehicle Activity 42
3.4.3.1. Engine Start Activity 42
3.4.3.2. Intra-zonal Running Exhaust Activity 43
3.4.3.3. Modal Activity 43
3.4.3.4. Road Grade 44
3.4.3.5. Temporal Variability 45
3.4.4. Facility Emissions 46
3.4.4.1. Engine Start Zonal Facility Estimates 47
3.4.4.2. Minor Road Zonal Facility Estimates 50
3.4.4.3. Lineal Facility Estimates 51
3.4.5. Emissions Inventory 55
3.5. CONCLUSION 56
4. MODEL DEVELOPMENT. ...57
4.1. INPUT FILES 60
4.1.1. Directory structure 60
4.1.2. Zone, twt and zip. twt 61
4.1.3. Allroads (ARC/INFO coverage) 61
4.1.4. Tdfh.dat (INFOfile) 62
4.1.5. Census (ARC/INFO coverage) 62
4.1.6. Landuse (ARC/INFO coverage) 63
4.1.7. ZIP code (ARC/INFO coverage) 63
4.1.8. TAZ (ARC/INFO coverage) 63
4.1.9. TAZ.dat (INFOfile) 63
4.1.10. Landmarks(ARC/INFO coverage) 64
4.1.11. Grid (ARC/INFO coverage) 64
4.1.12. Grade.xy and grade, gr 64
4.1.13. Lookup files 64
4.2. THE MAKEFILE 65
4.3. THE MODULES 66
4.3.1. Zonal Environment Module 67
4.3.2. Road Environment Module 67
4.3.3. Zonal Technology Groups Module 68
4.3.4. Major Road Technology Groups Module 68
4.3.5. Engine Start Activity Module 69
4.3.6. Major Road Running Exhaust Activity Module 69
' 3.7. Minor Road Activity Module 70
•1.3.8. Engine Start Emissions Module ; 70
4.3.9. Minor Road Running Exhaust Emissions -. 71
4.3.10. Major Rodd Running Exhaust Emissions : 71
4.3.11. Gridded Emissions ?7
IV
-------
.4.4. CONCLUSION 71
5. MODEL DEMONSTRATION 99
5.1. PREPROCESSING 101
5.7.7. Vehicle Characteristics 707
5.7.2. Conflation 702
5.7.3. Other Steps 702
5.2. SPATIAL ENVIRONMENT 103
5.2.7. Engine Start Polygons 104
5.2.2. Running Exhaust Lines and Polygons 705
5.3. FLEET CHARACTERISTICS 105
5.3.7. Model Year Distributions 705
5.3.2. High Emitters 706
5.4. VEHICLE ACTIVITY 106
5.4.7. Engine Start Activity 706
5.4.2. Running Exhaust Activity 706
5.5. FACILITY AND GRIDDED EMISSIONS 107
5.5.7. Engine Start Emissions 707
5.5.2. Minor Road Running Exhaust Emissions 108
5.5.3. Major Road Running Exhaust Emissions 108
5.5,4. SCF Running Exhaust Emissions 708
5.5.5. Total Emissions 709
5.6. CONCLUSION 109
6. MODEL EVALUATION 124
6.1. SPATIAL ENVIRONMENT 124
6.1.1.SZ 124
6.1.2. MR and MZ 726
6.2. VEHICLE CHARACTERISTICS 127
6.2.7. Zonal Fleet ...128
6.2.2. On-roadFleet 730
6.3. VEHICLE ACTIVITY 130
6.4. FACILITY AND GRIDDED EMISSIONS 131
6.4.1. Facility Emission Estimates / 32
6.4.2. GriddedEmissions 732
6.4.3. Sensitivity of Model 733
6.4.4. MEASUREvs. MOBILESa 141
6.4.5. Conclusion 148
1. ANTICIPATED CONTRIBUTIONS TO MOTOR VEHICLE EMISSIONS
ASSESSMENTS AND RECOMMENDED RESEARCH 149
7.1. IMPACTS 149
7.2. MAJOR CONTRIBUTIONS 150
7.2.7. Model Design and Development 750
7.2.2. Tool for the Exploration of Spatial Aggregation 7 52
7.2.3. Value of Geographic Information Systems 152
7.3. FUTURE RESEARCH 153
7.3.7. Model Validation Strategies 753
7.3.1.1. Spatial Environment 154
7.3.1.2. Fleet Characteristics 154
V
-------
7.3.1.3. Vehicle Activity „ .154
7.3.1.4. Facility and Gridded Emissions 155
7.3.2. Model Algorithm Improvement 755
7.3.3. Model Additions Research 756
8. REFERENCES ..... 157
APPENDIX A DATA DICTIONARY 162
VI
-------
TABLE OF FIGURES
FIGURE 1.1 - EMISSION MODELING SPECTRUM (TECH GROUPS REFER TO SETS OF VEHICLES WITH
SIMILAR EMISSION CHARACTERISTICS) 3
FIGURE 2.1 - CO EMISSIONS FOR A HYPOTHETICAL VEHICLE TRIP 10
FIGURE 3.1 - CONCEPTUAL MODEL DESIGN ......34
FIGURE 3.2 - SAMPLE REGRESSION TREE FOR NORMAL CO ENGINE STARTS (GRAMS/START) 40
FIGURE 3.3 - SPEED / ACCELERATION PROFILE, INTERSTATE RAMP, LOS D 44
FIGURE 3.4 - ENGINE START EMISSION PORTION 48
FIGURE 3.5 - RUNNING EXHAUST EMISSION PORTION ....51
FIGURE 4.1 - MODEL DESIGN ......59
FIGURE 4.2 - ZONAL ENVIRONMENT ENTITIES 73
FIGURE 4.3 - ZONALENV.AML FLOW CHART .....74
FIGURE 4.4 - ROAD ENVIRONMENT ENTITIES 75
FIGURE 4.5 - ROADENV.AML FLOW CHART 76
FIGURE 4.6 - VEHICLE CHARACTERISTIC ENTITIES 77
FIGURE 4.7 - TECHNOLOGY GROUP ENTITIES 78
FIGURE 4. 8 - ADDRESS MATCHING FLOW CHART 79
FIGURE 4. 9 - VEHICLES.MAK FLOW CHART 80
FIGURE 4.10 - 7£a/c/f.cFLOw CHART 81
FIGURE 4.11 -./O/ATG.C FLOW CHART 82
FIGURE 4.12 - ON-ROAD TECHNOLOGY GROUP ENTITIES 83
FIGURE 4.13 - MR-TG.AML FLOW CHART 84
FIGURE 4.14 - START ZONE ACTIVITY ENTITIES .85
FIGURE 4.15 - SZ-ACT.AML FLOW CHART 86
FIGURE 4.16 - MAJOR ROAD ACTIVITY ENTITIES .....87
FIGURE 4.17 - MR-ACT.AML FLOW CHART 88
FIGURE 4.18 - MINOR ROAD ACTIVITY ENTITIES 89
FIGURE 4.19 - MZ-ACT.AML FLOW CHART 90
FIGURE 4.20 - START ZONE EMISSIONS ENTITIES 91
FIGURE 4.21 - ES_EMISSION.C FLOW CHART 92
FIGURE 4.22 - MINOR ZONE ACTIVITY ENTITIES 93
FIGURE 4.23 - MZ-EM.AML FLOW CHART .94
FIGURE 4.24 - MAJOR ROAD EMISSIONS ENTITIES 95
FIGURE 4.25 - RE_EMISSIONS.C FLOW CHART 96
FIGURE 4.26 - GRIDDED EMISSIONS ENTITIES 97
FIGURE 4.27 - GRID-EM.AML FLOW CHART 98
FIGURE 5.1 - MODEL STUDY AREA SITE MAP 100
FIGURE 5.2 - ENGINE START ZONE CREATION 103
FIGURE 5.3- RUNNING EXHAUST ENTITY CREATION 104
FIGURE 5.4 - MODEL YEAR DISTRIBUTION 105
FIGURE 5.5 - HIGH EMITTER ENGINE STARTS, 7-8 AM 110
FIGURE 5.6 - HOME-BASED WORK TRIP TEMPORAL DISTRIBUTION 110
FIGURE 5.7- HOME-BASED SHOPPING TRIP TEMPORAL DISTRIBUTION 111
FIGURE 5.8 - HOME-BASED GRADE SCHOOL TRIP TEMPORAL DISTRIBUTION ill
FIGURE 5.9 - HOME-BASED UNIVERSITY TRIP TEMPORAL DISTRIBUTION „ '. 112
FIGURE 5.10 - NON-HOME-BASED TRIP TEMPORAL DISTRIBUTION 112
FIGURE 5.11 - ROAD VOLUME DENSITY 113
vn
-------
FIGURE 5.12 - ON-ROAD ACTIVITY TEMPORAL DISTRIBUTION Mv« -113
FIGURE 5.13 - ENGINE START CO, 7-8 AM J14
FIGURE 5.14 - MINOR ROAD RUNNING EXHAUST CO, 7-8 AM 115
FIGURE 5.15 - MAJOR ROAD RUNNING EXHAUST CO, 7-8 AM 116
FIGURE 5.16 - SCF RUNNING EXHAUST CO, 7-8 AM 117
FIGURE 5.17 - TOTAL CO, 7-8 AM 118
FIGURE 5.18 - TOTAL HC, 7-8 AM 119
FIGURE 5.19 - TOTAL NOX, 7-8 AM 120
FIGURE 5.20 - TOTAL CO, 6 AM - 9 PM ...121
FIGURE 5.21 TOTAL HC, 6 AM TO 9 PM 122
FIGURE 5.22 - TOTAL NOX, 6 AM TO 9 PM... 123
FIGURE 6.1 - POLYGON OVERLAY ERRORS... 125
FIGURE 6.2 - MODEL YEAR FREQUENCIES 129
FIGURE 6.3 - MODEL YEAR FRACTION 129
FIGURE 6.4 - OBSERVED ON-ROAD VEHICLE ORIGINS 130
FIGURE 6.5 - SAMPLE GRID CELL AGGREGATIONS... 133
FIGURE 6.6 - CO NORMAL EMITTER TECHNOLOGY GROUP EMISSION RATES BY LOS AND GRADE . 135
FIGURE 6.7 - CO HIGH EMITTER TECHNOLOGY GROUP EMISSION RATES BY LOS AND GRADE 136
FIGURE 6.8 - HC NORMAL EMITTER TECHNOLOGY GROUP EMISSION RATES BY LOS AND GRADE .137
FIGURE 6.9 - HC HIGH EMITTER TECHNOLOGY GROUP EMISSION RATES BY LOS AND GRADE 138
FIGURE 6.10 - NOX NORMAL EMITTER TECHNOLOGY GROUP EMISSION RATES BY LOS AND GRADE139
FIGURE 6.11 - NOx HIGH EMITTER TECHNOLOGY GROUP EMISSION RATES BY LOS AND GRADE... 140
FIGURE 6.12 - MEASURE G/SEC CO EMISSION RATES BY VELOCITY AND ACCELERATION FOR THE
STUDY AREA'S VEHICLE FLEET 142
FIGURE 6.13 - MOBILE5A G/SEC CO EMISSION RATES BY VELOCITY AND ACCELERATION FOR THE
STUDY AREA'S VEHICLE FLEET 143
FIGURE 6.14 - MEASURE G/SEC HC EMISSION RATES BY VELOCITY AND ACCELERATION FOR THE
STUDY AREA'S VEHICLE FLEET 144
FIGURE 6.15 - MOBILE5A G/SEC HC EMISSION RATES BY VELOCITY AND ACCELERATION FOR THE
SAMPLE AREA'S VEHICLE FLEET 145
FIGURE 6.16 - MEASURE G/SEC NOx EMISSION RATES BY VELOCITY AND ACCELERATION FOR THE
SAMPLE AREA'S VEHICLE FLEET 146
FIGURE 6.17 - MOBILE5A G/SEC NOX EMISSION RATES BY VELOCITY AND ACCELERATION FOR THE
SAMPLE AREA'S VEHICLE FLEET 147
Vlll
-------
GLOSSARY AND ACRONYMS
AQL - Georgia Tech's Air Quality Laboratory.
ARC - The Atlanta Regional Commission, the MPO for Atlanta, Georgia.
ARC/INFO UNIX Based GIS Software by Environmental Systems Research
Institute.
CBD - Atlanta's Central Business District.
Conflation - The process of transferring textual information from one linear data
representation to another.
Engine start - Term referring to the emission rate phenomenon occurring during the
first few minutes of a vehicle's operation.
Enrichment Term referring to the emission rate phenomenon occurring during high
power demand driving.
FTP Federal Test Procedure, the emission test cycle from which the MOBILE5a
emission rates were derived.
Geocoding The process of establishing locational parameters (coordinates) from
textual data.
GIS Geographic Information System, computer hardware and software used for
storing, displaying, analyzing, and modeling spatial information.
GPS Global Positioning System, a device used to determine one's position on the
earth's surface by triangulating distances from satellites.
HC - Hydrocarbons.
High emitters Term applied to a small portion of the fleet that produces higher
emission rates, usually the result of malfunctioning equipment.
IX
-------
Hot-stabilized - Term referring to the 'stable' emission rates characteristic of vehicles
operating with active emission control equipment, usually occurring after a vehicle
has warmed sufficiently.
LOS - Level of service is used to characterize operational conditions within a traffic
stream and their perception by motorists and passengers.
Makefile - A text script used to manage multiple programs and files.
MEASURE Mobile Emission Assessment System for Urban and Regional
Evaluation, the model developed by the research reported here.
MOBILESa - The active mandated emission rate model developed by the USEPA.
;i)U
Modal emissions - Emissions that have been separated by specific operating conditions
that result in distinct changes in emission rate behavior.
MPO - Metropolitan Planning Organization.
NAAQS - National Ambient Air Quality Standards, health-based air quality standards
that cities must not exceed.
Normal emitters - Term applied to vehicles with low to moderate emission rates due to
normal operation of emission control equipment.
NOx - Nitrogen oxides.
Photochemical models - Computer models used to predict ambient air quality.
Pollutants of concern Carbon monoxide, -hydrocarbons, and oxides of nitrogen.
Ozone Pollutant caused by the complex mixing process of NOX and HC in the
presence of sunlight.
Raster - Cell-based spatial data structure.
Running exhaust - Term applied to non-start exhaust pipe emissions that occur while a
vehicle is in operation.
SCF - Speed Correction Factor, the tech e found in MOBILESa for adjusting
emission rates based on the average spei t a vehicle, or sets of vehicles.
SOV - Single occupancy vehicle.
-------
Sub-fleet - Term applied to any group of vehicles smaller than a regional operating
fleet.
TAZ - The TAZ represents a spatial unit for aggregating socioeconomic data and
resulting trip generation estimates.
Technology group Term applied to categories of vehicles with similar characteristics
resulting in similar emission rates.
TIN - Triangulated Irregular Network.
TMIP - Travel Model Improvement Program, USDOT plan to improve the standard
travel demand forecasting modeling capabilities used by cities.
TRANPLAN Travel demand forecasting software produced by the Urban Analysis
Group.
TRANSIMS - Transportation Analysis and Simulation System.
Travel demand-forecasting models Models that follow the standard four-step
modeling strategy to predict travel behavior based on socioeconomic and
infrastructure data.
Unix - Unix is an operating system originally developed in the 1960's and 1970's by
scientists at the University of California at Berkeley and at AT&T Bell
Laboratories and was designed to be used for running scientific and engineering
applications on large processors.
USDOT - United States Department of Transportation.
USEPA - United States Environmental Protection Agency.
UTPS - Urban Transportation Planning System, a travel demand forecasting model,
developed in the 1960s.
Vector Topologic spatial data structure (points, lines, polygons).
VIN Vehicle Identification Number, a code number revealing many vehicle
characteristics and found on most vehicles.
XI
-------
1. INTRODUCTION
Suburban sprawl, population growth, and auto-dependency have, along with
other factors, been linked to air pollution problems in U.S metropolitan areas.
Accordingly, the Clean Air Act and other federal legislation and regulations require
metropolitan areas to develop strategies for reducing air pollution in those cases where
air quality standards are exceeded. An emissions 'budget' is established in these
metropolitan areas that provides a benchmark for comparing new emission-generating
activity, and presumably not exceeded. Such a goal becomes difficult when trying to
accommodate the needs of a growing population and economy while simultaneously
lowering or maintaining levels of ambient pollutants. Growing urban areas must,
therefore, continually develop creative strategies to curb increased pollutant
production. Because the largest contributor of pollutant emissions in urban areas has
most often come from transportation (or mobile) sources, transportation is targeted for
new control strategies.
Developing measures of effectiveness and subsequent predictions of overall
impact for control strategies require an understanding of the relationship between
observable transportation system characteristics and emission production. Quantifying
this effectiveness requires modeling these relationships. According to published
research, motor vehicle emission rates are correlated to a variety of vehicle
characteristics (weight, engine size, emission control equipment, etc.), operating
modes (idle, cruise, acceleration, and deceleration), and transportation system
conditions (road grade, pavement condition, etc.) [Guensler, 1994, Earth, 1996].
Exhaust emissions are produced when a vehicle is started and when it is in operation.
Pollutants produced from starting a vehicle can be predicted using vehicle
characteristics. Running exhaust emissions additionally require estimates of dynamic
engine conditions that result from how the vehicle is driven. Estimating motor vehicle
emissions requires the ability to predict or measure these parameters for an entire
region at a level of spatial and temporal aggregation fitting the scope of control
strategies. Current modeling approaches, however, do not have the capability to
provide these estimates.
Today's motor vehicle emission modeling process is based on four separate
models: a travel demand forecasting model, a mobile emission model, a
photochemical model (for emission inventory), and a microscale model (for analyzing
transportation improvements). The travel demand-forecasting model uses
characteristics of the transportation system and socioeconomic data to develop
-------
estimates of road-specific traffic volumes and average speeds. Mobile emission
models use these travel demand estimates, operating fleet model year distributions,
and environmental conditions to develop estimates of mobile source pollutant
production. These estimates are fed into photochemical models (along with stationary
source estimates and data regarding atmospheric conditions) and are used to predict
ambient pollutant levels in space and time. These mobile source estimates can also be
used by, microscale models to predict pollutant levels near specific transportation
facilities.
There are several problems with the four-model system that limit effective
(evaluation of motor, vehicle emission control strategies. First, the estimates of vehicle
activity (vehicle miles traveled and average speed) Jack', the accuracy and spatial
resolution needed to evaluate control measures [Stopher, 1993]. Second, the mobile
source emission rate modeling .process uses highly aggregate fleet.- estimates and
average emission rates which are not specific for the-fleet in operation, mode of
vehicle operation, or, grade of the highway facility. As.a consequence,'the current
modeling system has limited capabilities for meeting the modeling requirements of
transportation planners. Transportation planners and environmental assessment and
control officials have need for improved models that help identify the impacts of
standard transportation system improvements (e.g., lane additions, signal timing, peak-
hour smoothing).
While many researchers agree that new models and processes need to be
developed to overcome these problems, there is disagreement over the best approach
[Washington, 1996]. The U.S. Environmental Protection Agency and the Federal
Highway Administration held a workshop in Ann Arbor, Michigan'in May, 1997 for
the purpose of identifying and discussing current emission modeling research efforts
[Siwek, 1997]. After the workshop, it was clear that defining appropriate model
aggregation levels is important, in defining how and .what research should be
conducted. A point of departure between the largest vehicle emissions research efforts
(University,of California at Riverside, and the .Georgia Institute of Technology) and
..the.,currently^mandated approach (MOBILESa) isithe level of aggregation required.
Figure 1.1 demonstrates the,spectrum of possible approaches. The figure shows that
highly, aggregate approaches limit explanatory power, but have reduced data intensity.
Disaggregate models have the most explanatory power, but the highest data needs. An
added dimension to the issue is the fact that estimates must be spatially and temporally
resolved, suggesting that an undefined level of spatial and temporal aggregation must
also be defined; In fact, the level of spatial and temporal aggregation of mobile source
emissions needed by photochemical models may help define the minimum level of
model aggregation currently being debated.
-------
This report presents a research model that can guide future mobile emissions
model development efforts. A major objective of the model is to incorporate the latest
transportation / air quality findings at a low level of spatial aggregation (restricted only
by data availability). By creating a model under these guidelines, information is
developed that leads to the maximum level of disaggregation given user needs and
data availability. The research model will be comprehensive, flexible, and user-
oriented. It includes enhanced vehicle activity measures; starts, idle, cruise,
acceleration, and deceleration. Vehicle technology characteristics (model year, engine
size, etc.) and operating conditions (road grade, traffic flow, etc.) are developed at a
large scale (small zones and road segments). Flexibility is achieved through a modular
design that separates emission production based on thresholds determined in
background research. Due to large gaps in the state of knowledge, technology, and
practice regarding travel behavior, emission rates, and the urban system inventory, the
accuracy of the model results remains unvalidated and therefore unknown. However,
the model contributes to transportation and air quality research in that it aids research
and software development endeavors.
Emission Rates
Average
Fleet
(g/trip)
Vehicle Class
Average Speed
.(g/mi)
Tech Groups
Vehicle Mode
(g/sec)
Tech Groups
Vehicle Mode
(g/sec)
Aggregate
Total
Trips/Day
Vehicle Class
Tech Group
VMT Speed/ Accel.
Mean Link Speeds Profiles
Vehicle
Tech Group
Traffic Flow
Simulation
Activity
Individual
Vehicles
Vehicle Mode
(g/sec)
Individual
Vehicles
Engine Mode
(g/sec)
w
Disaggregate
Vehicle
Activity
Simulation
Vehicle/Engine
Activity
Simulation
Figure 1.1 - Emission Modeling Spectrum (tech groups refer to sets of vehicles
with similar emission characteristics)
The intended model users include emission science experts, model developers,
transportation planners, policy makers, and governmental researchers. Each user
group has specific modeling interests that define how the model should be designed
and presented. Central to the model design is a geographic information system (GIS).
Geographic information systems are widely used computer tools that allow
geographically referenced data to be organized and manipulated. Both transportation
and air quality vary in spatial dimensions. Thus, GISs have the conceptual capability
to manage the relationships between transportation activity and resulting air quality
-------
changes based on their spatial characteristics. Further, GISs are already used by most
planning organizations and government institutions. Thus, a GIS-based emissions
modeling .framework fits the character of emission science as well as fitting the
technical environment of the expected users.
The variables included in the proposed research model are those whose
relationship to vehicle activity and emission rates has been defined in research and
available to public agencies (see below). They can be categorized as follows:
Spatial Character:
• US Census block boundaries
• Land use boundaries
• Traffic analysis zone boundaries (from travel demand forecasting model)
• Grid cell boundaries (defined by user)
• Road segments (by classification)
• Travel demand forecasting network links
• Grade school and university locations
Temporal Character:
• Hour of the day
Vehicle Technology:
• Model year
• Engine size
• Vehicle weight
• Emission control equipment
• Fuel injection type
Modal Activity:
• Idle
• Cruise
• Acceleration
• Deceleration
Trip Generation:
• Home-based work trips
• Home-based shopping trips
• Home-based university trips
• Home-based grade school trips
• Home-based other trips
• Non-home-based trips
-------
Road Geometries:
• Number of lanes
• Grade
Socioeconomic characteristics (for spatial allocation only):
• Housing units
• Land use (residential, non-residential, and commercial)
1.1. Summary of Contributions to Research
• An automobile exhaust emissions model is developed maximizing
comprehensiveness, flexibility, and user friendliness.
Comprehensiveness is accomplished through the inclusion of variables and
procedures identified in the literature as significant to emission rate modeling.
Flexibility is achieved by organizing the model components by geographic location,
and by maintaining a modular program design. User friendliness is achieved by
including only current data available to planning agencies, and by using a GIS
framework.
• A research tool is provided that allows for the testing of variable levels of motor
vehicle emission model spatial aggregation.
By having the flexibility to use a variety of spatial entities, the model can
become a 'testbed' for determining the spatial resolution needed for future models.
This information is valuable in identifying future research needs, costs of emission
estimation, model development, maintenance, and operation. A question this model
could be used to help answer would be, "Given the current state of research, does a 1
sq. km aggregation of ozone precursors provide enough resolution to predict ozone
formation, or would a 4 sq. km aggregation be better?"
• The benefits of using GIS for emissions modeling are demonstrated.
GISs provide the ability to organize data by location, in turn providing the
capability to develop relationships with new or existing spatial datasets. This allows
for the development of creative alternatives to model construction and provides the
ability of prioritizing emission control strategies based on location.
-------
• Research and data needs for improved spatial and temporal emissions modeling
are identified.
A study of background research into emissions modeling coupled with an
analysis of data available in Atlanta will determine gaps in important emission-specific
variables. Further, a prioritization of the data needs based on balancing explanatory
power and cost will guide future model development.
1.2. Report Organization
Chapter 1 presents introductory discussion of the research, providing a list of:
significant contributions, modeling components, and modeling apprpach.
Chapter 2 discusses background research significant to automobile exhaust
emission modeling, vehicle activity modeling, and geographic information systems.
This chapter identifies a research foundation of knowledge that is used to develop
model parameters.
Chapter 3 presents a conceptual model design that serves as the foundation of
the research approach. Accuracy, comprehensiveness, user needs, and enterprise
awareness are important considerations in developing this conceptual model.
Chapter 4 provides a physical model structure that can be used as a research
tool. The model will reside in a UNIX operating system and use Make, the C
programming language, and ARC/INFO. A step-by-step guide to model use is also
provided in this chapter.
Chapters 5 and 6 analyze a model implementation for a 100 sq. km area in
Atlanta. Each module of the system is studied using sensitivity analysis, or through
comparison of observed data
Chapter 7 will discuss data needs and present final conclusions. An expanded
model diagram will demonstrate how future vehicle types and operating modes can be
added to the system.
Chapter 8 lists references cited in the report, and Appendix A is a data
dictionary.
-------
2. BACKGROUND
This background chapter will review the key literature related to emissions
modeling. Four general areas are reviewed: automobile exhaust emissions, emission
rate modeling, motor vehicle activity modeling, and geographic information systems
(GIS). The automobile exhaust emission section will focus on the cause and effect
relationships of vehicle operation and emission production. The emission rate
modeling section will focus on techniques used by different modeling approaches to
determine vehicle emission rates. The vehicle activity section will review and identify
techniques for developing estimates of emission-specific vehicle activity. The GIS
section will discuss issues surrounding spatial and temporal modeling, and review past
uses of GIS in the transportation and air quality arena.
2.1. Automobile Exhaust Emissions
This section discusses three topics that are important in motor vehicle exhaust
emissions: the major pollutants, the cause and characteristics of their production, and
the concept of modal emissions. Understanding these three is crucial to designing a
system that is focused on cause and effect relationships.
2.1.1. Exhaust Emission Pollutants
The Clean Air Act of 1970 identified six air pollutants of concern in the United
States: carbon monoxide (CO), hydrocarbons (HC), oxides of nitrogen (NOx), sulfur
dioxide (SO2), paniculate matter (PM-10), and lead (Pb). Recently, PM-2.5 was added
to this list. Nationally, in 1994, on-road vehicles were reported to contribute 62
percent of CO emissions, 42 percent of HC emissions, 32 percent of NOx emissions,
~5 percent of SO2, 19 percent of PM-10 (PM-2.5 was unreported), and 28 percent of
Pb [USEPA, 1995]. Carbon monoxide, hydrocarbons, and oxides of nitrogen are
pollutants prevalent in automobile exhaust (PM-10 is produced by diesel engines and
tire wear and Pb is being successfully reduced by its elimination from gasoline). For
the purposes of this research, the term 'emissions' will hereafter refer to CO, HC, and
NOx. All of the pollutants present health dangers to people, animals, and vegetation.
Ozone (Oj) is produced through a complex, series of chemical reactions that result
from pollutants (HC and NOx) mixing in the atmosphere in the presence of sunlight.
Generally, ozone concentrations are highest in urban centers and downwind of urban
-------
centers. Ozone has been observed to vary spatially in an urban area, and that the
production of ozone is theeresult of pollutants mixing in space and time. It is also
interesting to note that biogenic sources of HC contribute significantly to ozone
production. ;For example, in the southeast United States, eliminating all the
anthropogenic .(-man-made) sources of HC would still not result in passing federally
mandated ozone standards due to the levels of HC produced by biogenic (vegetation)
sources. [SOS, 1994].'«. This indicates that a NOx reduction policy would better serve
ozone reduction in the southeast [NRC, 1991 ].
On-road vehicles have been significant contributors to air pollution since the
1940s. The trends in new car emission rates of CO have shown significant
improvement over*the last thirty years. The improvements have been attributed to
legislatively induced emission controls for new vehicles (see section 2.1.2). The
actual transportation contribution to overall CO emissions, however, has not declined
at the same rate, due in part to the fact that the mobile emission controls are designed
to affect only a portion of'the engine operating mode and because per capita vehicle
miles of travel have increased. In fact, vehicle miles of travel (VMT), auto ownership,
person trips, and fraction of single occupant vehicles (SOV) have increased
disproportionally to population growth [Johnson, 1993, Meyer, 1997].
2.1.2: The Mechanics of Exhaust Emissions
In ideal combustion, oxygen and fuel (HC) are combusted and produce-
byproduct emissions of carbon dioxide (CO2) and water (H2O). Air, however,
contains nitrogen (Nz) among other chemicals, and combustion is always incomplete,
producing byproducts of HC, CO, oxygen (Oa), carbon dioxide (CO2), water (HaO),
and NOx [Heywood, 1988, Jacobs, 1990]. The air to fuel (a/f) ratipjs an important
factor in determining the quantity of pollutants produced by combustion. Generally,
rich fuel mixtures (low a/f ratios) produce high amounts of CO and HC because
combustion is incomplete. Lean fuel mixtures (high a/f ratios) will typically produce
higher amounts of NOx (especially during very hot, lean conditions) and lower
amounts of CO and HC because combustion is more complete. When considering
vehicle activity, high power demand (sharp accelerations, heavy loads, etc.) creates a
rich fuel mixture resulting in elevated CO and HC emission rates while NOx generally
decreases. At high speeds with low acceleration rates, a lean fuel mixture develops
which increases NOx emission rates [Heywood, 1988].
Car manufacturers design automobile engines to maximize fuel efficiency and
to comply with federal certification tests (Federal Test Procedure (FTP)), which means
balancing the a/f ratio (through computerized engine controls) to its most efficient
point (stoichiometry). However, car manufacturers also design automobile engines to
provide power to meet consumer demand. The certification tests do not cover the high
-------
speeds (maximum speed is 56.7 mph) and high accelerations (maximum acceleration
is 3.3 mph2/sec.) where rich and lean mixtures occur [Earth, 1996]. Therefore, all
automobiles are allowed to have inefficient combustion at the high ends of the speed /
acceleration spectrum in order to provide drivers with greater power on demand. New
test cycles would provide incentives for car manufacturers to reduce the designed
enrichment events resulting from power demand. This reduction could significantly
lower new car emission rates.
Vehicle technology has changed dramatically over the last thirty years and
great strides have been made in reducing emissions. In the 1960s, many vehicles were
fitted with devices that controlled the amount of fuel used for combustion, thereby
improving the efficiency of combustion and reducing exhaust emissions. In the late
1970s and early 1980s, catalytic converters were installed on new vehicles. Initially,
these catalytic converters focused on controlling CO and HC emissions. The catalytic
converters treated exhaust gas by removing much of the CO, HC, and NOx emissions
[CARB, 1990]. Because there is variability over time (model year) in the types of
emission control devices installed on new vehicles, it is probable that vehicle
characteristics will play an important role in predicting emission rates, and thus be an
important feature in model design for many years to come.
Because emission control technology significantly impacts emissions
generation, there are large differences between vehicles with functional control
systems, and those with malfunctioning, deteriorated, or nonexistent control systems.
The latter group can have significantly higher emissions [Pollack, 1992]. The
differences can be pronounced enough that researchers have termed the high emitting
vehicles 'high emitters.' Correct representation of high emitters in the vehicle fleet
will be crucial to accurate emission modeling efforts given the magnitude of these
"above normal" emissions.
2.1.3. Modal Emissions
Modal emissions refers to the types of emissions related to specific modes of
operation. Figure 2.1 conceptually represents the relative magnitudes of exhaust
emissions for a vehicle trip in space and time. As seen in the diagram, the initial rate
of emissions is high, indicating engine start mode. After the engine warms over a
period of time, emissions drop and stabilize (hot-stabilized mode). The stabilized rate
is interrupted by periods of high emissions (enrichment mode). Each of these three
automobile exhaust-operating modes is discussed in the following sections.
-------
Engii3<-S
taut
Acceleration
Enrichment
Engine
Rate
Engine On
Grade
Enrichment
Space and Time
Engine Off
Figure 2.1 - CO Emissions for a Hypothetical Vehicle Trip
2.1.3.1. Emissions in Start Mode
Motor vehicle emission rates are elevated during the first few minutes of
vehicle operation. This is primarily caused by emission control equipment that
functions well only at high temperatures. The magnitude of the emissions is a function
of: commanded air/fuel ratios, catalyst temperature, and engine temperature (Jacobs,
et al., 1990; Heywood, 1988; Joy, 1992; Pozniak, 1980). Most onboard computer
control systems initially demand an enriched fuel mixture so the engine will not stall
or hesitate during the warm-up period. Thus, the high emissions concentration in the
exhaust plume is initially a direct function of the computer control system which
varies from vehicle to vehicle. Commanded enrichment may cease when a specific
time has passed or when a specific coolant temperature is reached. As engine
temperatures rise, combustion efficiency improves and emissions concentrations are
gradually reduced. Finally, to be effective, catalytic converters must reach "light-off
temperatures of roughly 300 °C. Until the catalyst reaches this temperature, emission
concentrations in the exhaust plume remain high. Catalyst temperature rise is a
function of initial catalyst temperature, exhaust gas temperatures, exhaust gas volumes
passed through the converter, and emission concentrations.* "Thus, the magnitude of
elevated emi '. ns associated with engine -'arts is also a function of the amount of
time the verm:": has remained inactive (t!,.;; affects the catalyst and exhaust gas
temperatures), and a function of the manner u which the vehicle is operated after the
10
-------
engine is started (which affects exhaust gas volumes and hydrocarbon loading). Cold
starts, engine starts that occur when the engine temperature is below the catalyst light-
off threshold, have higher CO and HC emissions.
Two approaches have typically been employed to model engine start emissions:
1) starts are modeled as discrete emission-producing activity, or a "puff," and 2) starts
are modeled as a function of a base emission rate (hot-stabilized exhaust) adjusted for
conditions that elevate emission rates (Guensler, 1994). The California Air Resources
Board's,(CARE's) emission rate model (EMFAC7F), for example, treats the elevated
engine start emissions as a single "puff (i.e. separate from running exhaust) and
multiplies the number of engine starts by a cold start emission rate. The US
Environmental Protection Agency's emission rate model (MOBILESa), on the other
hand, increases the calculated running exhaust emission rate for vehicles, based upon
an assumed fraction of vehicles operating in cold start, hot start, and hot stabilized
modes. MOBILE5a documentation recommends using 20.6% as the percentage of
operating vehicles in cold start mode and 27.3% in hot start mode (based on the FTP
analysis). These percentages do not consider location or functional class and were
highly correlated to time of day and trip purpose [Venigalla, 1995a].
Historically, the number and location of cold starts have been based on trip
generation models (see section 2.3.1) using socioeconomic predictors. Considering
emission output, the major factor is not the actual number of starts, but the duration
and location of a vehicle operating in start mode. Therefore, a vehicle trip lasting
through the start mode will have significantly greater total pollutant production than
the few seconds of a false start (an engine start that does not result in a vehicle trip).
Research has shown that 180-240 seconds is the approximate average cold start mode
duration. In 200 seconds, a vehicle traveling at 35 mph can travel over two miles. A
spatially resolved model of start emissions must be able to identify the trip origin and
the point on a traveled route where a vehicle moves from elevated emissions in start
mode to reduced emissions in hot-stabilized mode. Given that the actual duration of
the start mode is not necessarily 200 seconds but a function of a number of engine
parameters and conditions, the ability to model on a large scale where the switch in
operating modes occurs for a fleet of operating vehicles becomes quite complex.
Because trip generation is estimated on a zonal basis, a zonal distribution of engine
parameters and conditions may provide enough regional disaggregation and zonal
aggregation to identify quantities of pollutants produced. Crucial to success, however,
is the size of the zone.
The determination of whether a start is "cold" or "warm" (a warm start occurs
when the engine is still warm and therefore closer to catalyst light-off temperature) is
also a difficult problem. The duration of the engine soak time (length of time the
vehicle is not running) has been used to determine whether a vehicle has a cold or
11
-------
warm engine, thus affeumg the duration of elevated .emission rates [Sabate, 1994].
Cold starts occur after 4 hours of engine-off activity for non-catalytic converter
vehicles, and after 1 hour for catalyst equipped vehicles. Therefore, the parking
duration of vehicles indicates how long it will take before the engines warms
sufficiently after a start.
The engine "cold" and "warm" start conditions pose a difficult modeling
problem. The temporal characteristics of vehicle start activity play an important role
in predicting appropriate emission rates. The travel patterns of vehicles also become
important. A model including cold and warm start vehicle activity must be spatially
and temporally resolved and include predictions of travel behavior and vehicle
technology descriptions."
2.1.3.2. Emissions in Hot-stabilized Mode
Hot-stabfl&ed emissions occur after a vehicle's engine has reached sufficient
catalyst light-off temperature. When the emission control equipment runs efficiently,
emission rates reach a low, fairly stable level. The stabilizing effect also occurs on
non-catalyst vehicles due to decreased commanded enrichment, cylinder quenching,
and engine oil viscosity. The stabilized emission rates actually fluctuate slightly
according to vehicle characteristics, environmental conditions, and vehicle operating
modes [Guensler, 1993a]. Vehicle characteristics that have been identified as possibly
having explanatory power for a vehicle's emission rate include model year, engine
size, accrued mileage, emission control equipment type (such as catalytic converter
type) and condition, fuel delivery technology, engine monitoring and control strategies
(integrated into the electronic control module), gear shift ratios, and vehicle weight
and shape (for aerodynamic drag) [Guensler, 1994, Earth, 1996]. Environmental
conditions include ambient temperature, altitude, and humidity [Guensler, 1994,
Earth,'1996]. Vehicle operating modes include cruise, acceleration, deceleration, idle,
and induced vehicle loads (e.g., number of passengers, trailer towing, grade, and air
conditioning) [Guensler, 1994, Earth, 1996]. A vehicle can move in and out of hot-
stabilized emission mode when sufficient power is demanded causing a rich air to fuel
ratio. When power is demanded causing an enriched fuel condition, emission rates
change dramatically (see section 2.1.3.3).
Current models account for some but not all of the factors listed above.
Instead, surrogate factors, which are correlated to the factors of interest, are used
because they are much easier to obtain for a regional fleet of vehicles. For example, in
the EPA MOBILESa mcur! an~ the California Air Resources Board EMFAC7F
models, the effects of . .-elen.uon, deceleration, cruise and idle are currently
represented by a single surrogate factor, average operating speed. Average operating
speed is correlated with different proportions of vehicle operating modes. Surrogate
12
-------
vehicle attributes include model year, fuel delivery technology, catalytic converter
type, accrued mileage, and vehicle condition, and are relatively easy to obtain or
estimate for a regional fleet of vehicles from registration and inspection / maintenance
databases.
2.1.3.3. Off-Cycle Exhaust Emissions
Off-cycle emissions are those emission events which occur outside the
envelope of the Federal Test Procedure (FTP). The FTP dynamometer test cycle was
used as the basis for current model emission rates. Because the FTP cycle did not
include vehicle activity with speeds above 57 mph and accelerations greater than 3.3
mph/sec, a certain portion of actual vehicle activity is unrepresented in the test dataset.
Activity outside the tested ranges would represent high engine loads and throttle
positions that push engines into enrichment conditions. These events are of crucial
importance, not just because they aren't included in the analysis of emission rates used
for current models, but because these events are known to produce the highest
emission rates [Benson, 1989; Groblicki, 1990; Calspan Corp., 1973; Kunselman, et
al., 1974]. hi fact, one sharp acceleration may cause as much pollution as does the
entire remaining trip [Carlock, 1993]. Emissions models may be underpredicting
emissions by fairly high margin.
Spatial modeling of off-cycle exhaust emissions requires the ability to predict
vehicle speeds and accelerations at a resolution deemed significant by emission rate
research. Speeds and accelerations could identify the fraction of the fleet that may be
unrepresented in current emission rates. Further, research into the reanalysis of second
by second emission test data is discovering substantial amounts of test data outside the
FTP envelope [Siwek, 1997]. The reanalysis could predict emission rates based on
speed and acceleration characteristics. Further, there is a need to develop emission
estimates at a facility level [Venigalla et al.,1995a]. That is, it must be able to predict
the locations of enrichment events. If facility-level speed and acceleration profiles can
be predicted, emission rates can be applied.
2.2. Automobile Exhaust Emission Rate Prediction
Three emission rate modeling approaches are discussed in this section; an
emission-factor approach, a physical approach, and a statistical approach. Each model
type has particular advantages and disadvantages. All of the approaches suffer from
two limiting factors:
13
-------
Inadequate. Vehicle Test Data. There is a significant amount of emission test data
compiled over the years (over 700 vehicles and over 8000 vehicle tests). Most
of the testing was done by agencies attempting to determine new car
conformity to emission standards. New cars are run through the Federal Test
Procedure (FTP) which is a set of three^test cycles run on a dynamometer.
There is a cold start cycle (bag 1), a running exhaust cycle (bag 2) and a hot
start cycle (bag 3). The cycles are called 'bag' data because emissions are
collected in a bag during the test. All of the test datasets suffer from at least
one of two major limitations, sample size and (or) unrepresentative cycles.
The FTP cycle, for example, does not test accelerations above 3.3 mph/sec or
speeds above 57.5 mph. Other test cycles that have high speed and
acceleration data do not have a representative sample of the on-road fleet.
Inadequate prediction of emission-specific vehicle activity. Emission-specific
vehicle activity refers to the division of vehicle operation into groups that differ
significantly in their resulting emission rates. All of the approaches require
vehicle activity as an input. The best predictor of vehicle activity for
metropolitan areas is currently the four-step Urban Transportation Planning
System (UTPS) (see section 2.3.1). Although advances have improved the
ability of these models to predict emission-specific vehicle activity, most
MPOs still use models that have significant errors in facility-level estimates of
volume and average speed [Stopher, 1993, Harvey, 1991, Cutwater, 1994].
All of the modeling approaches focus on developing emission production
estimates, but few present systems are designed to address facility-specific impact
issues. This issue is crucial in defining which emission rate model approach best fits
the technical capabilities and economic constraints of agencies required to make
estimates. In other words, the most accurate model for predicting the emissions of an
individual vehicle may not be the most useful for certain types of modeling situations.
It may also become evident that the understanding of the causes of an individual
vehicle's emission rate has greatly surpassed the ability to collect the input variables
for a real-world operating fleet. Important to this issue is the level of aggregation
manifested in deterministic or stochastic approaches.
2.2.1. A Speed Correction Factor Approach
Both ' ;? USEPA's a^d the CARB's modeling systems use a 'speed-correction
factor' approach to predict gregate emission rates. The svstems are mandated for
use in conformity deterni; on despite their widespread atistical and theoretical
criticism. The models seleci a base emission rate dependir g on a variety of vehicle
technology and environmental parameters. The base emission rate is then factored or
adjusted based on the ratio of the observed speed to the average FTP cycle bag 2 speed
14
-------
(1-6 mph). As the models are currently used, the documentation suggests using default
values for national fleet averages and other variables. On the positive side, the
modeling system is not data intensive, it requires only inputs of total vehicle miles
traveled (VMT), average speed, and a cursory knowledge of fuel type and climate data
to get estimates of pollutant production. The system is easy to use and widely-
implemented by agencies without significant capital or operating expense. On the
negative side, it is not responsive to changes in important variables (acceleration, fleet
makeup, engine load, etc.).
The emission rate models are based on data collected from the FTP cycles
developed for new car emission testing. Added to the problems noted earlier with the
FTP cycle data, the modeling methodology is highly aggregate and therefore
insensitive to microscale variability [Guensler et al., 1993b]. The approach, therefore,
may not be able to accurately identify the best choice between small scale
development alternatives (changes in lane widths, signal coordination, etc.).
The EPA's Office of Mobile Sources continues to support research which will
help to identify incremental improvements to their modeling process. Currently,
MOBILE 5a is the mandated emission rate model, and MOBILE 6 is under
development. Modal issues, non-FTP cycle estimates, and other emission rate specific
factors are planned for implementation.
2.2.2. A Physical Approach
The physical or deterministic approach to emissions modeling is designed to
develop accurate emission estimates using many variables. The University of
California at Riverside is currently developing such a modal modeling approach under
a three year National Cooperative Highway Research Program project. The approach
will track the vehicle components and conditions that affect emission rates. The
model is designed to track an individual vehicle's power demand and engine
equipment status. Power demand is predicted using environmental parameters (wind
resistance, road grade, air density, temperature, and altitude), and vehicle parameters
(velocity, acceleration, vehicle mass, cross-sectional area, aerodynamics, vehicle
accessory load, transmission efficiency, and drive-train efficiency). Power demand is
combined with other engine parameters (gear selection, air/fuel ratio, emission control
equipment, and temperature) to develop dynamic vehicle or technology group
emission rates. When combined with a vehicle's operating parameters, deterioration
(the change in emission rate over time due to catalyst decay or equipment
malfunction), and fuel type, the model produces highly time resolved emission
estimates which promise to be more accurate at the microscale level than any model
produced thus far. Vehicle test data for their model are being collected on
15
-------
dynamometers (-300 vehicles) as part of the project. Final test data should be
available in two to three years [Earth 1996].
Barth, et al. recognize that their approach is data intensive, but accurate
emissions modeling forces it to be so. The vehicle data requirements are many and go
beyond the availability of information found in vehicle identification numbers (VINs)
that are maintained by state registration datasets. A lookup table could be developed
for missing parameters based on vehicle make, model, and model year. Other data
(environmental, and operating conditions) would have to be developed from other
models. The physical approach fits well with a simulation model of vehicle activity
(see section 2.3.2) because the simulations track individual vehicles.
The use of the physical model approach for regional impact modeling requires
data aggregation. As with other models, the approach is plagued by poor estimates of
emission-specific vehicle activity. Because the physical approach appears to be the
most accurate model for predicting an individual vehicle's second-by-second emission
rate, vehicle-specific second-by-second activities are needed to get accurate results.
Because accurate prediction of these parameters relies on predicting human behavior
among other highly variable data, it is likely the activity estimates will have high
variability. Aggregating to statistical distributions of the data will lessen this problem,
but departs from the original intention of highly accurate second-by-second estimates.
The large number of input variables introduce error associated with the ability in
predicting their values. The algorithms may be solid, but data input error could
significantly degrade the accuracy of the final estimate.
2.2.3. A Statistical Approach
Researchers at Georgia Tech have developed a modeling approach that is based
on statistical distributions of a variety of vehick technologies and vehicle operating
modes. The' core of the emission rate mode is based on hierarchical tree-based
regression analysis (HTBR). The tree analysis is a statistical procedure that iteratively
splits a dataset into two parts by; (1) selecting a variable that controls the most
variability, and (2) determining a cutpoint of that variable that explains the most
variability. The result is a 'tree' where each ending node is a set of predictor variable
conditions, and an emission rate (for each pollutant and operating mode). Once a
'tree' is developed, adjustments are made to the values based on load (from wind
resistance, grade, and accessories).
Georgia Tech researchers combined a variety of emission test datasets from a
number of sources in order to maximize the comprehensiveness of the vehicle fleet
and potential operating conditions. The data have been re-analyzed to allow modal
parameters to be included. Although there are still limitations with the dataset
16
-------
(representative fleet and cycle operating conditions), an extensive emission rate 'tree'
has been developed. The HTBR approach is also plagued with the lack of availability
of adequate data input. Extensive vehicle data (model year, engine size, fuel system,
emission control, vehicle class, vehicle test weight) and vehicle operating data (speed,
acceleration) are needed for predicting emissions. One benefit to the approach is that
it can be adjusted, for missing data. If one particular variable is missing from the
dataset (vehicle test weight, for example, is not stored in the Vehicle Identification
Number), the HTBR can be re-run and produce new emission rates that exclude that
variable. The new rate may, however, be less accurate, depending on how significant
the missing variable is to emission estimation. Another benefit to the statistical
approach over current models is the ability to put confidence bounds around each
estimate. This becomes important when estimates for a variety of conditions on a
certain facility segment are added together to produce a single facility estimate, whose
accuracy must be quantified.
Critics of this modeling approach have suggested that the inability to track
causal variables results in a model that is unable to predict the effects of new
technology. There are three counter-arguments to this criticism, (1) because control
standards continue to tighten, it is more important to model the old technology instead
of the new, (2) no model can expect to accurately predict future technology changes,
they can only develop relationships based on known conditions, and (3) if surrogate
variables are correlated to casual ones, the model will still continue to work.
2.2.4. Emission Rate Modeling Summary
The microscopic physical approach taken by Earth et al. has the potential to
provide the most explanatory power, disregarding input data error issues that can't be
quantified at this time. It is also clear, from the research that the speed correction
factor approach is highly aggregate and inappropriate for the modeling needs of
research and planning agencies. The statistical approach provides near-term
improvements and allows for facility level aggregations of data. An important factor
in selecting a particular emission rate modeling approach is its ability to fit within the
framework of the larger 'data model' issues regarding the user needs of measuring and
predicting transportation impacts on air quality. The 'data model' in this context
refers to the design of an entire modeling system from user needs to data structure and
connectivity. The statistical approach seems most appropriate given the scope of this
research because it appears to fit the balance between accuracy and implementability
identified as a modeling objective in Chapter 1.
17
-------
2.3. Vehicle Activity Modeling
2.3.1. Urban Transportation Planning System (UTPS)
The Urban Transportation Planning System, (or travel demand forecasting
model), first developed in. the 1.960s, was designed to predict travel flows within an
urban area. The primary purpose of the system was to guide new infrastructure
investment [Cutwater, 1994J. Because of its predictive nature and widespread use, the
use of this modeling approach has expanded beyond the original intent to predicting
emission-specific vehicle activity. Until recently, vehicle activity has meant vehicle
miles traveled (VMT) and average speed, the inputs to mandated emission models.
However, as understanding of emission behavior expands, so does the definition of
vehicle activity. Emission-specific vehicle activity now encompasses detailed modal
parameters which UTPS models are incapable of predicting. Researchers have
identified numerous deficiencies in the approach (outside implementation problems);
the facility level (link) estimates are highly variable, the models do not predict off-
peak travel well, seasonal variations in travel are not considered, model size is limited,
and the models are not sensitive enough to measure mandated TCM effectiveness
[Stopher, 1993, Harvey, 1991, Cutwater, 1994].
Along with theoretical problems, there have been a significant number of
implementation problems including: lack of feedback components, insufficient current
socioeconomic data and inadequate validation procedures [Harvey, 1991, Cutwater,
1994]. Model results have indicated an accuracy range of 5-30% error in overall VMT
estimates and 5-20 mph error in average speeds. [Miller, 1995]. Average error by
models implemented by MPOs is 10% for VMT and 15 mph for average speed
[Stopher, 1993]. Errors also increase as one moves from higher to lower road
classifications. To add to the problems, the same models that are criticized as too
simplistic are too complicated and costly for proper implementation by many agencies.
Despite these errors and theoretical deficiencies, the models represent the state-
of-the-practice. In fact, they represent the only short and medium range alternative
available for widespread implementation. There is a significant amount of research on
techniques for improving the UTPS and hopefully improvements will result in better
predictions of vehicle activity in time and space.
The Travel Model Improvement Program administered by the US Department
of Transportation is attempting to improve the travel forecasting capabilities. Some of
the potential improvements to predicting emission-specific vehicle activity are as
follows. (1) There is a shift away from trip-based motu towards activity-based
models. Activity-based models better represent term :ul changes and mode
alternatives. (2) Development of stochastic microsimulation techniques aggregated to
18
-------
area traffic patterns will allow improved sensitivity to temporal changes. (3) The use
of longitudinal panel surveys will more accurately identify cross-sectional survey
(current technique) biases.
2.3.2. Simulation Models (TRANSIMS)
Simulation models are being viewed by many as the solution to the problems
facing the UTPS. Simulation models generally come in three forms, microscopic,
mesoscopic, and macroscopic. Microscopic models track individual vehicles and their
relationships with other vehicles. Macroscopic models approximate traffic flow as a
fluid and use a facility (road segment) as a the base unit. Mesoscopic models combine
elements of both depending on the needed function. Simulation models have
successfully been used for optimization (signal timing, traffic flow) and for forecasting
(predicting results of a change). Models can be deterministic or stochastic (by
allowing some randomness into the process). By their nature, simulation models have
the theoretical and computational capability to predict facility-level activity at a
resolution needed to predict emission-specific activity. The structure and data
requirements of existing models have prevented their implementation for an entire
urban structure, and force use at the facility level. Most models have been developed
to answer specific problems instead of complete system simulation. However, a new
generation of simulation models is taking a broader scope and the models are being
designed around regional systems instead of specific traffic-flow issues. Recent
advances in modeling theory, microscopic modeling, and computing power may have
expanded the role of traffic micro-simulation modeling from the facility scale to the
urban/regional scale.
Advances made by "TRANSIMS" (Transportation Analysis and SIMulation
System) have led many to believe that they have found a replacement for the UTPS
type models. TRANSIMS is being developed under the US Department of
Transportation's Travel Model Improvement Program and funded by the Federal
Highway Administration and other federal agencies. The intent of the project is to
develop a system that will be able to answer questions regarding policy and
infrastructure change for an entire urban area. One of their major selling points is their
focus on predicting air quality and other environmental impacts.
TRANSIMS will be a set of modules that can be run separately or together.
The first module is a household and commercial activity module that uses US Census
data to develop a synthetic population of individuals for Census Tracts and Block
Groups and predicts synthetic economic activity and resulting travel demand. The
second module is the intermodal route planner that takes the activity-based travel
demand and develops trip plans for every individual that can be adjusted depending on
the activities of other individuals over time. The third module is the travel'
19
-------
microsimulation module that tracks . individuals and their vehicles, and their
relationships to other vehicle activities, on the road network using a 'cellular
automata' technique. The final module is the environmental module that predicts a
variety of environmental conditions including mobile source emission prediction,
atmospheric mixing, and concentrations. The outputs of TRANSIMS will be
summaries of second by second data at cells of 7.5 meters.
TRANSIMS promises to provide unique solutions to the integration of
macroscopic and microscopic transportation modeling and provide advances in a
number of simulation issues. Issues that the developers must address are validation
and implementation. All of the new algorithms and techniques must be individually
validated against observed data. The time frame and cost of implementation at a new
urban area may be extensive due to the input data requirements.
Despite a number of issues that must be addressed by TRANSIMS developers,
it is apparent that the spatial and temporal resolution of emission-specific vehicle
activity could be substantially improved by TRANSIMS in the future. This aspect
identifies the emission modeling need for incremental research that builds towards a
future system that can move toward the objectives defined by the Los Alamos
researchers.
2.4. Geographic Information Systems
A geographic information system (GIS) is "a computer-based information
system that enables capture, modeling, manipulation, retrieval, analysis and
presentation of geographically referenced data" [Worboys, 1995]. The rise of GIS
technology and its use in a wide range of disciplines provides transportation and air
quality modelers with a powerful tool for developing new analy••?., capability. The
organization of data by location allows data from a variety of sources to be easily
combined in a uniform framework. For example, vehicle registration information can
be combined with census data to develop driver-vehicle profiles. Or, high traffic
volume areas can be combined with satellite analysis of vegetation decay to study
environmental impacts. Another important feature of GIS is its ability to bridge the
technical gap between analysts' and decision-makers' need for easy-to-understand
information . The communication power of GIS (thematic maps, GUIs, 3-D surface
plots, etc.) is a feature th_ as made GIS one of the most used platforms for planning
in the U.S. GIS provide, the ability to get quick answers to technical questions.
Literature on GIS data structures, applications, and vendor products is substantial. The
following section will briefly cover, 1) the extent of GIS implementation by
20
-------
transportation and air quality agencies, and the past use of GIS in transportation and air
quality analysis, and 2) the issue of spatial data quality.
2.4.1. GIS in the Transportation / Air Quality Agencies
The National Cooperative Highway Research Program (NCHRP) Report 359
studied GIS in an effort to define its potential for transportation agencies. The
document, which presented a comprehensive overview of GIS technology, its potential
role for serving the needs of a variety of agencies, and strategies for successful
implementation, stated,
" The potential impact of GIS-T is profound. If this technology is
exploited to its fullest, it will become ubiquitous throughout all
transportation agencies and will become an integral part of their
everyday information processing environments. ... The potential impact
of GIS is more than just agency wide. The problems of today require the
interaction of agencies at all levels of government. ... the broad
problems that are driving the interaction typically involve environmental
and economic development issues; and their solutions will require the
integration and analysis of geographically referenced data of many kinds
from many sources."
2.4.2. Applications of GIS in Mobile Emission Modeling
2.4.2.1. Emission Inventories
Bruckman et al. presented a paper in 1991 at an Air and Waste Management
Association conference describing the use of GIS in developing gridded, hourly
estimates of emissions. They also developed a model called CAL-MoVEM that
utilized GIS in developing mobile source estimates for input into photochemical
models. The main function of the GIS in their model was the spatial aggregation .of
travel demand forecasting model features into a grid. They used spatially defined
vehicle mixes by trip purpose, temporal factors, hourly temperatures, trip volumes, trip
speeds, and modal percentages as inputs. The spatially defined inputs were combined
with EMFAC7E emission rates to produce gridded hourly estimates of pollutants
[Bruckman, 1991]. The work was accomplished as part of a study on ozone levels in
the San Joaquin Valley in California. Zonal estimates were allocated to TAZ (traffic
analysis zone) centroids that were re-allocated to grid cells. Link estimates were
allocated to nodes and re-allocated to cells. The use of points to represent these
features did not take full advantage of the spatial structure provided by the original
input data. TAZs falling along grid cell boundaries should have their portions divided.
This strategy would limit grid cell sizes to those significantly larger than TAZs, which
21
-------
can be quite large (30-40 square km) for some metropolitan areas. Also, no mention is
was made of strategies for identifying the confidence ranges of the estimates.
The model supports the use of GIS, but did not take full advantage of the
research value of GIS. Further, the model did not have the flexibility to answer the
diverse impact or mitigation questions that arise from estimating emissions.
2.4.2.2. GIS for Transportation Planning and Air Quality Analysis
juai,
Researchers used GIS as^-p^eprocessor and postprocessor to mobile emission
modeling. Although they relied on existing models to estimate emissions, they
showed how GIS could be valuable in the management of emission related data. They
made the connection between the needs of transportation planners and decision-
makers and the spatial tools and features of GIS {Souleyrette 1991].
2.4.2.3. Microscale Analysis
Researchers at Utah State University used GIS in, developing microscale
analyses of a small group of intersections. They linked a G^S with CALINES, and
CAL3QHC to predict pollutant concentration levels [Hallmark, 1996]. The value of
GIS (outside of spatial data storage and data visualization) was its ability to compare
concentration results to other non-related data. The contribution is significant to this
research because it provides a foundation for the argument that a GIS approach is not-
restricted to developing emission inventories, but can be easily expanded to a number
of other related issues.
2.4.2.4. Influencing Decision-makers
Othofer et al. developed an interesting approach to predicting location specific
emission production estimates for changing control strategies. Instead of developing
estimates using detailed location-specific emission producing activities and emission
rates, they disaggregated large zonal estimates using emission-producing activities
[Orthofer, 1995]. The advantage of this approach is its simplicity and its
straightforward recognition that the data needed to predict emissions at smaller levels
does not exist or the relationships are undefined. The disadvantage is that the ability
to predict changes among the disaggregated levels is a function only of the change of
the overall larger units. Thus, the true effects of activity changes on emissions cannot
be measured. The prou produced high-quality graphics that indicated locational
variation in emission-proc. cmg activities. The project was successful because elected
officials could 'see' areas that have potentially high emissions and therefore had
evidence for developing actions for those specific areas. Although, the modeling
22
-------
capability of the project is limited, its ability to influence action through spatial
communication is a noteworthy contribution to the use of GIS in this arena.
2.4.3. Spatial Data Issues
Spatial data .refers to points, lines, or polygons that maintain a digital
connectivity with other entities in regards to their relative position. Spatial data comes
in two forms, raster or vector. Raster data is information in a regular unit, usually a
grid cell. The grid cell maintains an attribute value and locational information
pertaining to its place in a matrix. Raster data is preferred when representing
continuous data (natural features, environmental features, air quality, etc.) or when
developing complex spatial data models. Vector data is information in the form of
points, lines, or polygons. Vector data better represents features with discrete edges
(anthropogenic features, rivers, transportation, etc.). An issue of prime importance to
both data structures, and for this research, is spatial data quality. The quality refers to
a number of issues regarding the accuracy and resolution which are discussed in the
next two sections.
2.4.3.1. Positional Accuracy
Positional accuracy refers to the variability of the represented position from the
actual position. Relative positional accuracy refers to the relational position between
represented features and absolute position accuracy refers to the relationship between
represented features and the Earth's surface. A good relative accuracy and poor
absolute accuracy indicate a positional problem that is important when bringing
different databases together. Because of the development of US National Map
Accuracy Standards and the advent of improved surveying techniques, relative
positional accuracy within a single dataset is not significant given the scope of
inventory modeling. Absolute positional accuracy becomes an issue when joining
multiple spatial databases. Variations in position can result from using different
projections, datums, or transformations. Any attempt to join spatial databases must
address the issue and provide solutions (stretching, fuzzy tolerance, etc.) to reduce the
impacts.
2.4.3.2. Data Resolution
Data resolution concerns the level of spatial aggregation, or density of
observed values. Data resolution usually refers to the scale at which the original data
observations were made, and the level of interpolation used in developing the final
dataset. The level of resolution is important in determining the confidence of a
represented value at a particular coordinate. As in positional accuracy, data resolution
problems usually occur when trying to combine databases of varying resolution. The
23
-------
combination of two datasets will result in a dataset that has a resolution equivalent to
the one with less detail. This is frequently overlooked in analysis resulting in the
presentation of data with significant variance. For example, soils data at a scale of
1:24,000 can be overlaid with 1:100 parcel data in an attempt to identify the parcel's
soil type. The result represents the 'best guess' as to soil type variations within a
parcel, but the variability is high. It is good practice to question whether the scale of
database fits the spatial character of what is being represented. For example, does a 1
km or 4 km aggregation of ozone precursor pollutants provide enough resolution given
the scale of ozone formation?
2.4.3.3. Data Content Accuracy
Data content accuracy refers to the accuracy of the attribute data represented by
the spatial feature. Data content accuracy can be limited by a number of procedural
problems (coding error, measurement error, etc.) or by the change of the data over
time. Data content can be estimated by validation techniques, but they are usually
cost-prohibitive for the large spatial datasets available. Usually spatial databases can
be tracked to an original collection technique that may have been validated. It is also
possible to compare two or more datasets for agreement to develop a qualitative
appraisal. Most publicly available datasets have quantitative information on the
accuracy of their data content.
24
-------
3. MODEL CONCEPTUAL DESIGN
This chapter presents a conceptual design of the GIS-based automobile exhaust
emissions model. The background research from the previous chapter is summarized
into a series of 'research foundation points' that define modeling parameters. User
requirements are also identified, guiding model form and presentation. By the end of
the chapter, a modeling approach is recommended.
While the overall purpose of the model is defined in Chapter 1, more specific
model objectives that guide the development of such a model include:
• The model must produce automobile exhaust emission estimates that are capable
of being statistically verified.
It is vital for the model to be able to determine errors in estimates that result
from input data error and algorithm error. One of the biggest criticisms of the
currently mandated modeling approach is that there is no information available for
users to estimate errors resulting from the algorithms. A design open to outside review
and analysis prevents avoidable extrapolation because the confidence intervals are
known.
• All estimates and input parameters (emissions, vehicle activity, etc.) must be
capable of being validated.
All model components must be capable of being validated either through
previously, published research or through designed experiments. Given the
complicated process of predicting emissions, it is important that all intermediate
modeling steps be designed to be tested. This objective will influence the data model
because many elements regarding vehicle technology and vehicle activity will have to
have identifiable characteristics that can be observed in the field.
• The model must be designed to easily incorporate new findings.
Because research into emissions modeling is occurring in a number of
institutions, significant findings are expected in the near future. Keeping the research
model up to date to research from other institutions is crucial if it is going to be used to
influence research direction and software development decisions.
• The model must use available data.
Although the model is not designed to be implemented on a widescale basis for
official reporting, it must still be constrained by real-world conditions of data
25
-------
availability and cost. Without considering these factors, one can spend a significant
amount of time and resources developing models from variables (dynamic engine
parameters, etc.) that cannot be collected by a regional modeling agency.
• The model must use as large a spatial scale as data will allow.
It is important to use available data, but it is also important to use the largest
scale possible. One of the uses of the model will be to identify the level of spatial
aggregation required for useful emission estimation. In order to do this, it is important
to start with the most detail and aggregate up, thereby identifying locations with high
emission production.
• The model must be portable.
The model should be transferable to other urban areas without substantial
model alteration. This means that all the input parameters should be available to
major metropolitan areas, and that model assumptions should not be limited to the
study area.
3.1. Model Design Parameters
The following model design parameters are based on material discussed in the
background chapter, user requirements, and good modeling practice. These
parameters will establish minimally acceptable guidelines for model development.
The ability of the model design to abide by the parameters will depend on the data and
technology available to a clearly-defined user group. Some parameters may have to be
scaled back due to limitations in data availability.
This summary of the background knowledge is presented in this section to
clearly identify model development parameters. The initial goal of model
development is to include all listed parameters. At some point limited data availability
or excessive data development expense will likely remove or scale back some
parameters from consideration. The research backed parameters are:
• Develop estimates of the production of automobile exhaust pollutants CO, HC, and
NOx in space and time (from section 2. J.I)
Research has shown tuat the major exhaust pollutants of concern are CO, HC,
and NOx. Considerations should be given to including paniculate matter greater than
2.5 microns in diameter (PM2.5) due to its recent identification as a health risk.
However, there is very little data on the cause and effect relationships of PM2.5
production by automobiles.- The emission estimates represent only the production of
pollutants, not the resulting air quality. The spatial and temporal scale should be
26
-------
developed according to anticipated user needs. Existing photochemical models (used
to predict ambient air quality) currently use hourly, 4-5 sq. km aggregations. Future
photochemical model improvements are expected to use 1 sq. km estimates of mobile
sources.
• Anthropogenic NOx estimate accuracy important in predicting ground-level ozone
(from section 2.1.1)
Major cities in wanner - climates have air quality problems resulting from
ground level ozone concentrations. NOx and HC are precursors to ozone formation.
HC, however, can be produced in significant amounts by biogenic sources. Therefore,
a more accurate, verifiable, estimate of NOx may prove more useful in predicting the
impact of motor vehicles.
• Comprehensive representation of vehicle technologies (from section 2.1.2)
Differences in vehicle technologies / characteristics have been shown to
significantly affect vehicle emission rates. As seen in the physical model approach by
Earth, et al., it is actually the dynamic status of a number of vehicle parameters that
causes emission rate variability (see section 2.2.2). At the same time, a number of
vehicle characteristics have been tied to emission rate variability because they are
surrogate variables for causal parameters (see section 2.2.3). From a research model
perspective, it is important to be able to include both sets of conditions. However,
modeling individual vehicle engine dynamics for an urban area is not practical due to
extensive data requirements. Instead, only those specific static inventory variables
involved with the dynamic conditions, and those variables identified as surrogate
variables are included. The list of desired vehicle characteristics are: model year,
engine size, weight (or mass), emission control type(s), fuel delivery type,
transmission type, cross-sectional area, and number of cylinders.
• Separate and quantify high-emitting vehicle emissions (from section 2.1.2)
A small percentage of the fleet disproportionally contributes to total mobile
source emissions. By separating this small high-emitting portion of the operating fleet,
it will be easier to predict the impacts of control strategies that may target high
emitters. Further, model attention should be focused on factors that result in higher
emissions, wisely using resources in the most important areas.
• Separate start, hot-stabilized, and enrichment emission quantities and locations
(from section 2.1.3)
By separating estimates into specific emission modes, mode-specific impact
strategies can be more efficiently evaluated. Further, emission rates for each mode are
predicted using different variables. Engine starts are primarily influenced by vehicle
characteristics and engine temperature. Hot-stabilized and enrichment emissions are
-------
primarily influenced by vehicle characteristics and operating condition (speed,
acceleration, etc.).
• Include Speed Correction Factor (SCF) emission rates (from section 2.2.1)
The inclusion of SCF emission rates provides an alternative modeling
approach. One of the objectives of this research is to be-comprehensive and flexible.
Inclusion of the SCF estimate provides a flexible framework and a way to compare
between emission rate modeling approaches. The model may indicate that the highly
aggregate SCF approach is suitable for regional inventory modeling at a certain spatial
level. The SCF approach to modeling start emissions will not be included because the
approach lacks the ability to show spatial variability between start and running
emissions.
• Include emission rates from the statistical approach (from section 2.2.3)
Emission rates from the statistical approach need to be included because the
research indicates that modal parameters better characterize accurate emission rate
estimation. Because the modal emission rates models are available, they can be
immediately integrated into the model framework. The approach also produces
separate start and running exhaust-emission estimates, addressing one of the previously
defined model design parameters.
• Include activity measures from travel demand forecasting models (from section
2.3.1)
Travel demand forecasting models are the primary predictive tool for regional
level vehicle activity. They are also used by almost every transportation planning
agency (MPO) in the country. Further, their use in developing emission estimates is
currently mandated. Despite their well-documented problems, they have
characteristics that make them very attractive for a spatially-resolved model. First of
all, they have a defined structure and connectivity that translates into a spatial form
(zones, links, and nodes). Second, they develop estimates using socioeconomic
information, allowing the model to be indirectly affected by social and economic
changes.
• Prepare for inputs from future simulation models (from section 2.3.2)
Simulation models provide vehicle activity measures at a larger spatial scale
and resolution than macroscopic travel demand models. The value of producing
estimates a: the microscopic or mesoscopic scale becomes evident when studying the
types of vehicle activity that produce high emissions. Just as high emitters
disproportionally contribute to total emissions, so do high power demand situations.
These driving situations can be characterized by comprehensive representation of
traffic flow dynamics.
• Use geographic information systems (from section 2.4)
28
-------
Using GIS is important because it is designed to handle the spatial data
management and modeling functions key to the research goals. Without GIS, complex
spatial analysis and manipulation algorithms would have to be re-created. Its
widespread use and popularity among planning agencies is significant enough to
warrant its use.
3.2. User Requirements
In designing any analysis model, it is crucial to clearly understand the analysis
needs of the proposed users. There are several user groups that could be expected to
interact with the research model.
• Emission Science Experts
These experts are those individuals who help define the emission science
domain. They provide the knowledge regarding the cause and effect relationships in
automobile emissions modeling. Although their interaction with model design is
conceptual, it is tremendously beneficial if they can interact with specific model
components to ensure that the science is being accurately represented. Therefore, one
data model requirement is that the system be composed of well-documented and
appropriately termed modules that can be easily reviewed by the specific component's
experts. The model vocabulary should be defined by the experts' terminology (i.e.,
transportation components use standard traffic engineering terminology).
• Model Developers
Model developers can also have significant knowledge of the cause and effect
relationships among key variables. If a model requires significant software
development, it would be prudent to organize it using standard programming
techniques, terminology, and comments. This may require that comments in the code
explain the underlying scientific concepts to the point that clear understanding of the
importance of the various pieces is evident to developers. If a developer could
improve program efficiency by slightly altering a process, it would be beneficial that
the explanatory cost of the change be evident. Therefore, well-documented code is a
specific data model requirement.
• Emission Researchers
Emission researchers are individuals who use the model to get a better
understanding of the impacts of new findings, or develop criteria for future research
efforts. This adds a dimension to the model by requiring that measures of confidence
be included with the estimates. The estimates produced by the model must be'capable
of being accompanied with certain measures of accuracy and descriptions that clearly
29
-------
identify what is known and unknown in the process. The ;model inputs and outputs
should be hi a format that allows easy import/export to various software packages that
may be used for more detailed analysis. Outputs must include detailed summaries of
assumptions and discussions of accuracy to prevent false conclusions from being
drawn.
• Government Experts
Government experts would be individuals who would look at all levels of
model development to ensure quality and accuracy in order to approve or disapprove
results that could hold legal bearing. If model results are to be used for conformity or
inventory reporting, the model elements must be validated and peer-reviewed. This is
important in a developing model because government experts and researchers must be
included in the design process to prevent efforts from moving in directions that
contradict policies and mandates that govern air quality modeling. This user
requirement strengthens the need for modular, clearly communicated model code.
• Transportation and Environmental Planners
Transportation and environmental planners are the eventual 'users' of the
system. They will be the ones who develop the emission estimates for their particular
project. Although the level of development discussed in this report is for a 'research-
grade' model not to be used for legal reporting, the intention is that the model or some
of its components eventually be targeted for widespread public use. By including the
eventual user needs in the early design, complications in future development can be
avoided. By including transportation and environmental professionals, who may or
may not have model development experience, the system design becomes intuitive and
flexible. Planners should not be burdened with extensive command and syntax
requirements. Results should be designed towards the reporting needs of the planners.
• Non-technical Decision-Makers
Decision-makers (policy-makers, managers, planning boards, etc.) need model
results to make informed decisions. Decisions range from guiding the direction of
research to broad-based policy analysis and to local transportation alternative analyses.
Many of these users are removed from the modeling process, but must be familiar with
the process of modeling so they can have confidence in the model results and be aware
of assumptions made by modelers. By maintaining the model framework within an
off-the-shelf GIS, questions about model inputs and outputs can be asked and
answered by non-technical users. Further, thematic maps and user-def<;jd spatial
queries, graphical results can be produced along with standard spreadsheet and textual
reports.
In summary, the user-defined needs include:
• Appropriate documentation
30
-------
Appropriate terminology
Modular system design
Open input and output data formats
Intuitive modeling process
Easy to understand and use
Model should reside in a CIS
3.3. The Spatial Data Model
A good data model has five dimensions [Reingruber, et al., 1994]: conceptual
correctness, conceptual completeness, syntactic correctness, syntactic completeness,
and enterprise awareness. Conceptual correctness refers to the degree to which the
model represents the real world, or the accuracy of the model estimates. For this
model, it refers to the accurate representation of the cause and effect relationship
between motor vehicle behavior and emissions. Conceptual completeness refers to the
wholeness of the represented science. In this model, it refers to the ability to represent
the cause and effect relationships in a comprehensive manner for the entire urban area.
Syntactic correctness and completeness refer to the quality of the use of language and
proper communication. This would involve the use of organized and structured
programming techniques as well as the use of accepted transportation, air quality, and
GIS terminology. Enterprise awareness refers to the idea that the model does not
work in a vacuum, but that it represents only a portion of a much larger system. An
automobile exhaust model is a portion of the much larger scope of environmental and
transportation modeling. Keeping the model open to connectivity with other systems
ensures an adaptable and open system.
As the design of model components develops, it will be analyzed in respect to
these five concepts, paying particular attention to completeness and correctness.
These measures will guarantee that the design will have an organized framework.
Spatial data entities are the spatial forms used to characterize an object. For
example, a road 'object' can be characterized by the digital representation of a line, the
spatial data entity. Generally, the best entity type to use for spatial modeling or
environmental data representation is a raster cell. A raster cell structure handles
continuous variables better than a vector (points, lines, and polygons) structure. This
occurs because regular grid cells that fall between observed values can have
statistically interpolated values. A vector representation forces observed values to be
discrete within its structure, possibly misrepresenting boundaries. Another possible
spatial data entity is a triangulated irregular network (TIN). TINs interpolate points
found on a line between two observation points based on the values of the points. It is
31
-------
most often used in representing topography, but the concept could be translated to
other areas. The selection of an appropriate spatial structure is controlled by the model
objectives, model parameters, anduser needs.
A vector approach has the following advantages:
• Intermediate estimates must be validated
Because the model being developed is research-oriented, all of the algorithms
and data must be represented in a format that can be field validated. A clearly defined
beginning and ending point for a segment of road (intersection to intersection), or a
field-evident boundary (zone bounded by roads) makes validation simpler.
• Users require facility-level estimates
Transportation modelers (eventual users of the model) work with vector-based
facility entities (see section 3.4.1). By providing and receiving data in a similar
format, the integration and transfer of data is more efficient.
• Pollutant production is discrete
Vehicles are discrete objects. Because the model predicts pollutant production,
not resulting air quality, the emission estimate should also be discrete. Given that the
model will not actually model individual vehicles, there is an argument that an
aggregate characterization of factors is more efficiently handled in a raster approach.
However, the appropriate level of aggregation is undefined; in fact, one of the model
objectives is to provide a way to determine the appropriate level of aggregation. Once
it is determined, a raster approach for software development may be warranted.
A raster approach to the model also has benefits associated with its use in the
research model:
• Inventory estimates are gridded
The final outputs of mobile emissions models are inputs into photochemical
models. The photochemical models are raster due to the nature of the phenomenon
being modeled (ambient air quality). Gridded, hourly estimates are currently required.
• Data will have to be aggregated eventually
It is unlikely that widespread data availability allows modeling on a vehicle per
vehicle basis, nor is it likely that modeling at that level is practical or useful for
regional scale modeling. Some level of aggregation will have to be used. In that
regard, it may become more appropriate to predict continuous distributions rather than
discrete polygon values.
Overall, if the user is concerned with the location of high auto emissions, a
raster approach would be better, although technically questionable for linear features
32
-------
such as roads. If the user is concerned with the emissions on a specific entity (i.e.,
road segment, TAZ), a vector approach would be better. Given that the specified users
are concerned with both issues, both raster and vector entities should be used. The
issues of validation and emission science suggest that initially, vector data models are
warranted. At some point, the vector structure needs to be converted to raster for
further photochemical modeling, and regional data visualization.
3.4. Model Approach
The conceptual design of the proposed research model meets all of the stated
goals and objectives that have been stated earlier while avoiding the constraints of
extensive data development and cost. The model will be deterministic and spatially
and temporally identify: the types of vehicles that are being operated, the types of
activities the vehicles are involved in, the resulting emission rates, and the resulting
emissions. The level of aggregation and spatial scale is flexible, depending upon the
user's needs, data availability, and accuracy requirements of post-processing.
Regardless of the spatial scale, the conceptual design remains the same. Figure 3.1
shows a schematic drawing of the design. The top row represents the spatial
environment. The second represents vehicle characteristic assessment. The third
represents the vehicle activity. The fourth and fifth rows represent facility-level and:
gridded emission estimation. Detailed information about each component is provided
in chapter 4.
Central to the model design is the identification of the source of vehicle
activity data. While travel demand forecasting models have significant limitations
providing inputs to emissions modeling, they represent the only widely used
prognostic planning tool available. Until regional micro-simulation models become
widely accepted, validated, and implemented, emission models must rely on the
forecasting capabilities of the tools in use. With all of its disadvantages, there are
components of the travel demand forecasting model that defend its use for emissions
modeling. First of all, trip generation results can be easily translated to engine starts,
an important emission activity. Second, poor estimates of average speed can be
supplemented with observed speed and acceleration data given certain traffic flow
parameters. Third, the travel models have spatial characteristics that can form a
foundation for spatial modeling. The research model is tied to traditional travel
demand models. At some point in the future, regional simulation models may become
the primary source of travel behavior prediction. This should be reflected in the model
design by avoiding strategies that lock into specific travel model types.
33
-------
Spatial Environment
Ml
;, .i- ,. ,1
Off-Network
Module (polygons) '
K f * 4
Road Network
* Module (lines) '
: - ! *"
/ \ ... J.
Vehicle Characteristics
; ': . , '• •.-!,.'' '.'.' - "'•" >•: .
• •:• • ^Off-Network '>('•..? tr:$
Module ; c : .
;. \/. ;i, On-Network, ,'- " V
•• ••.. '.v -: , .Module .:. .•/.•C;!-". :
'• /;-';-'' 'v'". , ':": "':.:'• , >
Vehicle Activity
/ ""'•" : Off-Network '--^l-.' '•>'.'.
- Activity Module "!, . -l:
•A-f ' *i '-t, ••'':'trv"''t?-:^'ir;' ' ,"'.' i^ii
i . ' , On-Network v ',;....
"Activity Module
Facility Emissions
Off-Network ' ;.;'/;,
Emissions Module : •
- ' - On-Network :•.'•'/•; •. V,
Emissions Module '•
Mobile Emissions
Inventory Module
(Gridded, Hourly)
Figure 3.1 - Conceptual Model Design
The following sections describe the five major tiers of the model design. In
each section, descriptions of the roles, data needs, and processes are provided. Three
of the five parts of a good data model are discussed for each major component:
conceptual correctness, conceptual completeness, and enterprise awareness (the other
two parts are related to syntax and deemed less significant). Each section is
supplemented with specific model descriptions in chapter 4.
-------
3.4.1. Spatial Environment
The objective of the spatial environment tier is to unify input data under a
common zonal and lineal structure. The size and scope of the zones and lines depend
on the users and their specific needs. Historically, exhaust emissions are divided into
start and non-start (running exhaust) emissions. Most prognostic travel models
provide a zonal (TAZ) estimate of the number of trip origins, and a lineal (link)
estimate of road volume and average speed. By defining an engine start as being
synonymous with a trip origin, TAZs become the base spatial entity used for engine
start emissions. Running exhaust emissions occur on the road network, suggesting
that a 'link' should be the base spatial entity. Improvements in the spatial resolution
of the zonal estimates can be made outside the travel model by disaggregating trips to
smaller zones. The lineal estimate can similarly be improved by conflating (see
section 3.4.1.3) the links to comprehensive and accurate road datasets. These issues
are discussed further in the next two sections.
3.4.1.1. Zonal Data
The zonal module defines the polygon structure used to represent data and
emission estimates for engine starts. It is the role of the zonal module to combine the
polygons of various input data (i.e. socioeconomic, land use, TAZ) into a single
polygon dataset. As mentioned before, TAZs represent the base spatial entity for
engine starts. However, disaggregating trip origin estimates from large TAZs to
smaller zones can be accomplished if good socioeconomic and land use data are
available. For example, home to work trip origins can be assumed to start from the
residential areas within the TAZ. Likewise, return trip origins can be assumed to start
from land uses representing workplaces. While the process of disaggregation is
discussed later, it is the role of the zonal module to establish the data linkages that
make it possible.
Due to the fact that polygon data usually come from a variety of original
sources and therefore a variety of spatial representation differences, significant errors
may occur when trying to bring the datasets into a unified structure. It is unlikely that
boundaries that represent identical features from different datasets will match
perfectly. The result of this problem is the creation of a series of 'sliver' polygons
whose attributes may be misaligned. However, there is no loss of information, only a
zone structure that is as spatially accurate as the original data. The model does not
make any assessment about the spatial accuracy, but uses whatever data are available.
This allows users to define their spatial accuracy needs through the accuracy of the
input data. Thus, if one wants estimates of engine start pollutant production within
100 meters, one must provide input data with equal or better spatial accuracy than 100
meters. Each of the new polygons maintains key fields tying them to their original
35
-------
datasets, allowing all engine start emission estimates to be aggregated to any of the
input polygon structures.
3.4.1.2. Lineal Data
The road module defines the lineal data used for predicting running exhaust
emissions. While the travel demand forecasting models continue to be criticized for
inaccurate roadway volume and speed estimates, they represent the only available
prognostic regional vehicle activity tool. Most of the models predict travel (volume
and average speed) only for major roads, aggregating minor roads to TAZ 'centroid
connectors'. Further, the lineal representations of the road networks are usually
spatially abstract structures. 'Links' represent actual road segments, but given
modeling tasks, detailed shape points are unnecessary. In Atlanta, the absolute spatial
errors resulting from the abstract representation exceeded 2 km in some instances
[Bachman, 1996]. Improving this error is important to generating emissions for grid
cells of 4 sq. km or smaller.
3.4.1.3. Conflation
Conflation is the blending of two line databases. Conflating the abstract travel
demand forecasting network and a spatially accurate comprehensive road database is
needed to improve the spatial accuracy of the travel model results. Because travel
models' abstract 'links' represent actual road segments, it is possible to assess the
connections to other road dataset lines based on link configuration and attributes. The
process requires a link by link assessment and conflation by the user, resulting in a
time-consuming and tedious task. Many planning organizations that develop travel
demand forecasting models have already conflated the networks for purposes outside
emission modeling. Conflation is required for the research model and not considered
to be a task beyond the users needs. The model design can function without it, but at a
significant loss of spatial error, thereby eliminating one advantage of the approach.
The purpose of the model's road module is to separate the conflated road
dataset into modeled and unmodeled roads. Modeled roads, usually the roads with the
most volume, become the major lineal structure used to represent running exhaust
emissions. Unmodeled roads are aggregated into zones (bounded by modeled roads)
used to represent minor running exhaust emissions. An argument that supports the
zonal representation is the assumption that half the vehicles traveling on minor roads
have a higher chance of operating under start conditions because they are closer to
their origin (start conditions can last 2-3 minutes).
36
-------
• Spatial Environment Conceptual Correctness
The spatial environment modules have the task of defining the locational
parameters for the rest of the model. The structure of the spatial environment needs to
reflect the spatial characteristics of automobile exhaust emission production.
Automobile exhaust emissions are produced by operating vehicles traversing a road
network. The road network becomes the crucial component of the spatial
environment. For major roads, there is little loss in the conceptual correctness of the
spatial representation. Minor roads, however, suffer from insufficient prognostic data
forcing zonal aggregation. The zonal aggregation uses discrete polygons in
representing urban information. While the conceptual correctness suffers, important
linkages become straightforward and the needs for minor road modal activity are
lessened.
• Spatial Environment Conceptual Completeness
The conceptual completeness of the spatial environment refers to the
comprehensiveness of the spatial representation. The zonal aggregation of all roads
not modeled by the travel demand forecasting model ensures comprehensive spatial
representation by being a 'catch-all' for minor roads. The age of the input data will
impact the completeness of the data. Recent land use changes or new road
construction will be left out unless the input data are continuously maintained.
• Spatial Environment Enterprise Awareness
The spatial environment structure is based on zonal and lineal representations
of spatial structures used in a variety of agencies. By maintaining connections to the
original input dataset identifiers, solid linkages to these agencies are provided.
Further, by using a GIS and organizing data based on location, an indirect linkage to
many enterprises is possible.
3.4.2. Fleet Characteristics
Although different, emission modeling approaches are being developed, all
research efforts indicate that an improved capability to identify the emission
significant components of the operating fleet is important to emission rate accuracy
[Siwek, 1997]. Currently, emission models use model year distributions to describe
the fleet. However, many other vehicle characteristics hold significant explanatory
capability for predicting emission rates [Guensler 1994, Earth 1996]. Further,
spatially variant emission estimates are needed, requiring spatially resolved sub-fleet
characterization [Bachman, 1996]. Therefore, there is a need for identifying
procedures that can accurately predict spatially resolved vehicle characteristics for
urban areas. The fleet characteristics modules described in this section develop
emission-specific and location-specific estimates of the distribution of automobiles.
37
-------
-nagmml vahirln mQiftrntian .data-provide jafi iun Ifrir nlliu •imiiiiii.iiF
Impuitam rthiete chaiauuisrics td tie"tiaeiaanertat HfffivfatftrvehwMB. -The flats
nlso provide clues to identifying the vehicle-'* location, the WWMCT*S wgwiwwi addfcssr, *
and tht owiMfVZB^^ode. The fleet characteristics tier will develop estimates of
technology distributions for each of the zonal and lineal representations. There are
four general tasks: (1) attaching location parameters to the individual vehicle
registration data; (2) determining important characteristics of the vehicles; (3)
determining technology groups for the vehicles, and (4); aggregating to spatial entity-
specific technology distributions.
The first vehicle characteristics module has two major tasks: determine
individual vehicle location parameters and emission-specific characteristics. Each task
is time-consuming due to the size of the registration datasets found in a metropolitan
area. In Atlanta in 1995, 2.2 million vehicles were registered in the nonattainment
area. The initial intense processing tasks need to be completed only one time per year,
following new registration database development. Therefore," "the "first vehiete
'"CharacteTlsticb inudtrfe becomes a 'pm-piw^ccing*. .crg^. r**iAi*£ «MU«^u.tiM» formal
model.
3.4.2.1. Vehicle Geocoding
Address geocoding is a process whereby standard address fields of road name,
road type, and ZIP code are used to identify corresponding lines in a road database.
The address number is used to identify the position of the address on a matched line
based on the left and right address ranges. Address-matching usually results in success
ranges of 60-80% dependent on the quality and comprehensiveness of the road dataset,
and the number of errors associated with miscoding, duplicate or multiple road names,
apartment numbers, and rural route identification. Growing urban areas have difficulty
keeping road datasets current with new housing developments, adding significant bias
to the geocoding errors.
The geocoding process results in two types of records: matched and
unmatched. The matched vehicles are associated with a point entity. The unmatched
vehicles maintain a default location identifier of ZIP code, a polygon entity. While
ZIP codes can be rather large, they provide a degree of spatial information that can
help determine regional fleet distribution variability.
3.4.2.2. VIN Decoding
Raw registration data can usually provide a few important vehicle
characteristics (VIN, make, model, model year, and number of cylinders), but more
information can be developed from the vehicle identification number (VIN). All
38
-------
vehicles after 1980 are given a 17 digit VIN that consists of a code containing
information about the types of emission control systems, the fuel delivery systems, the
engine size, etc. Prior to 1980, VINs existed but lacked universal standards.
Decoding the VIN for each vehicle requires the use of software (VIN decoder)
developed by Radian International Corporation. Missing vehicle characteristics and
the lack of updates prevent sole reliance on Radian's VIN decoder [Bachman, 1998].
Missing characteristics, pre-1972 autos, and post-1994 autos need to be developed
from lookup files using the make, model, and model year. Research efforts at Georgia
Tech have resulted in a datafile that can be used to determine the test weight of
vehicles. While significant errors are expected, enough information should be
available to develop a clear view of the operating fleet distributions.
The vehicle characteristics module results in two groups of vehicles and their
emission specific characteristics; point-based (successfully matched) and zone-based
(unmatched). These files should represent a comprehensive description of the region's
fleet characteristics. These files are further processed to develop the emission-rate
specific fleet distributions.
The zonal technology group module takes the spatially-resolved vehicle
characteristics' files and determines zone-based engine start and running exhaust
technology group distributions. The technology group (TG) definitions are defined by
the emission rate modeling approaches included in the system. Currently, they are the
aggregate modal approach (see section 2.2.3) and the speed correction factor approach
(see section 2.2.1) because they are the only currently available models.
3.4.2.3. High and Normal Emitters
The aggregate modal approach developed emission-specific technology groups
using a regression-tree analysis of emission test vehicles [Wolf, 1998]. In the analysis,
all vehicles are divided into technology classes, indicating high or normal emitter
fraction likelihoods. High and normal technology groups are then defined for each
pollutant and emission mode (engine start and running exhaust). A sample engine
start, normal emitter, CO, 'tree' is provided in figure 3.2. By starting at the top of the
tree, conditions are identified based on the vehicles characteristics. True statements
move to the left side of the tree, false statements move to the right. Each ending node
is a set of conditions that are assigned a grams per start emission rate.
39
-------
MY<80.5
INEF TIA<3250
65
M/<79.5
D<198
240
90
110
M/<86.5
INEPT 'IA<3687.5
CATGONV<3.5 CATOONV<2.5
FIMJ<2.5
51
25
170
64
41 22
Figure 3.2 - Sample Regression Tree for Normal CO Engine Starts (grams/start)
A high emitting vehicle is one that has malfunctioning or tampered with
emission control systems causing higher than normal emissions. It is expected that a
small percentage of high emitting vehicles account for a large percentage of total
emissions. High emitter determination is an important model design parameter and
therefore it is appropriate to characterize these vehicles differently. The fraction of
high emitters in the fleet, and the rate of malfunction among different vehicle types are
unknown, but currently being researched. The regression-tree results by Wolf et al.
divide the fleet into four groups that have different likelihoods of being high, emitters.
Lacking better information, a random sample for each group will be separated and
labeled as high emitters, with sample sizes based on the group's likelihood. All other
vehicles will be modeled as normal emitters.
3.4.2.4. Technology Groups
Once vehicles are identified as high or normal emitters, they are characterized
into technology groups. Technology groups are combinations of vehicle
characteristics and operating conditions that have been identified in the regression tree
analysis as having significant emission rate differences. There will be separate
technology rroups for high and normal emitters, each pollutant, and each operating
mode. Eacn vehicle will fall into six technology groups (engine start CO, HC, and
40
-------
NOx, and running exhaust CO, HC, and NOx). Specific technology group
descriptions are provided in chapter 4.
Engine start technology groups only include vehicle characteristics. For each
emission-significant combination of vehicle characteristics, an associated gram per
start emission rate is identified. Running exhaust technology groups include vehicle
characteristics and(or) modal operating parameters (idle, cruise, acceleration, etc.).
Unlike engine start groups, running exhaust technology groups can have different
emission rates based on modal operating conditions. By the end of the fleet
characteristics' modules, distributions of technology groups will exist for every zone
in the model.
• Fleet Characteristics Conceptual Correctness
The representation of vehicles into emission-specific high and normal emitter
technology groups is based on observed relationships discovered through test datasets.
The ability of the technology groups to correctly represent the emission-specific
characteristics of the operating fleet directly relates to the representativeness of the
emission test dataset. Clearly this is not the case [see section 2.2]. However,
alternative conceptual approaches suffer from the same limiting factors. As new
vehicle tests are performed, and as re-analysis of past vehicle tests continues, progress
towards a representative fleet will be accomplished. In fact, the technology group
approach provides greatest potential for correct representation when representative
samples are not provided.
• Fleet Characteristics Module Conceptual Completeness
The 'conceptual completeness' of the vehicle characteristics approach is fairly
good, all operating automobiles are considered. However, data limitations severely
hamper comprehensive development. By using a region's entire registration dataset, a
comprehensive view of the region's vehicle characteristics is possible. Geocoding
errors and decoding errors result in a significant loss in data [Bachman, 1998].
Problems with completeness are resolved by developing distributions of technology
groups instead of frequencies. While lost data have bias and cannot be fully
represented, the use of distributions provides a 'best guess' given the data limitations.
• Fleet Characteristics Enterprise Awareness
The vehicle characteristics can be tied to other users of spatially-resolved fleet
descriptions because the locational parameters are defined before the fleet is
segmented into emission-specific technology groups. This allows the individual
vehicle characteristics to be available for other analyses. Inclusion of technology
groups from two separate modeling approaches directly results in added flexibility and
openness for the users.
41
-------
3.4.3. Vehicle Activity
As mentioned previously, the core prognostic capability of the model rests on
the ability of travel demand forecasting models to accurately predict regional travel.
The emission-important vehicle activity estimates provided by the regional travel
models are: the number and location of peak hour (or daily) trip origins, road segment
volumes, and road segment average speeds. Important activity not provided by current
models are; temporal travel behavior and modal (idle, cruise, acceleration, and
deceleration) operations. As indicated in the background chapter, the average speed
estimates can be very poor. Therefore, it is the role of the vehicle activity modules to
transfer usable travel model information into the modeling environment, and develop
estimates of the missing important parameters.
3.4.3.1. Engine Start Activity
Engine starts are equivalent to trip origins determined by the trip generation
component of travel demand forecasting models. Travel demand forecasting models
divide an urban region into traffic analysis zones (TAZs). The TAZs represent a
spatial unit for aggregating socioeconomic data and resulting trip generation estimates.
The designation of a TAZ should be based on homogeneous socioeconomic
characteristics, reducing the variability of the trip estimates. However, many urban
areas use zonal definitions based on cadastral (property) boundaries or US Census
boundaries. TAZs are usually large (2-5 sq. km) due to the original objectives of the
travel demand models (major infrastructure investments). Unless the TAZs can be
disaggregated to smaller zones, the TAZ structure will determine the spatial resolution.
: Estimates of trip generation are made for each TAZ for a variety of trip
purposes. Trip purposes usually include trip production and attraction estimates of
home-based work (to and from the workplace), home-based shopping, home-based
school, home-based other, and non-home based. While these trips are estimated to
begin or end in certain TAZs, the trip type definitions imply that they can be tied to
land use. For example, a home-based-work trip consists of a trip originating from
home (residential) going to work (non-residential), or a trip originating from work
going home. Likewise, home-based shopping trips imply trips to or from a
commercial land use.
The US Census Bureau maintains zonal databases developed for the decennial
census. The smallest zonal designation is a block, usually an area bounded by roads or
other lineal features (cadastral, hydrologic. etc.). At the census block level, 1990
estimates of the number of households are available. While the estimates are out-of-
date, they can possibly provide clues to housing density within the TAZ and land use
designations. This information can be used to further spatially disaggregate trips
originating from residential areas.
42
-------
By having good land use data and socioeconomic data, various trips can be
disaggregated to smaller zones. Even if the land use designations are as broad as
"residential" and "non-residential", the spatial resolution of trip generation estimates
can be improved, allowing an improved spatial resolution for engine start estimates.
3.4.3.2. Infra-zonal Running Exhaust Activity
The road network used by travel demand forecasting models usually consists of
major roads only. Travel on other roads is either not considered or predicted on an
aggregate zonal (TAZ) basis. A key variable in predicting running exhaust emissions
is the amount of travel time (preferably broken down by operating mode) because the
longer a vehicle is operating, the more pollutant is produced. Travel times for intra-
zonal trips (and inter-zonal travel off the major roads) are unaccounted for, other than
looking at the size of the zone. However, information exists that allows the
development of travel time estimates using the previously mentioned disaggregate trip
generation estimates, a digital road network, and spatial analysis tools provided by the
geographic information systems (GIS).
Many GISs provide tools that allow the determination of the shortest network
path between two points. The disaggregated trip generation estimates provide a trip
origin location. The closest intersection of local roads and major roads provides a
destination location, representing the point during the vehicle trip when the travel
demand models have assigned trips to the network. The shortest network path
between the two points provides an estimate of the travel distance. Averaging all the
distances within a TAZ provides an estimate of the typical intra-zonal and inter-zonal
travel distance that occurs before vehicles reach the modeled network. Assuming an
average speed for the local road travel provides an estimate of the average travel time.
Although, the strategy described above is crude and unvalidated, the method is better
than the alternatives of leaving the estimates out, or assuming travel times based on
zone area.
3.4.3.3. Modal Activity
Modal activity is a vehicle activity characterized by cruise, idle, acceleration or
deceleration operation. Research has clearly identified that modal operation is a better
indicator of emission rates than average speed [see section 2.2]. Determining regional
modal operation is not possible using current travel demand forecasting models alone.
All travel models can provide is road volume (± 15%) and average speed (± 30%).
Because the accuracy of the average speed is poor, it should not be used in emission
rate evaluation. However, the average speed could be accurate enough to determine
differences in levels of service (LOS) E and F where volume to capacity (v/c) ratios
approach or surpass 1.
43
-------
Research by Grant et al. is attempting to characterize speed and acceleration
profiles (Watson plots) by collecting data on major roads around Atlanta with a Laser
Rangefmder [Grant, 1996]. The research has produced results for freeway and ramp
sections by grade, LOS, and vehicle type. Using these results, speed and acceleration
profiles can be identified for prevailing conditions predicted by the travel demand
forecasting model. An example of a speed / acceleration profile is provided in Figure
3.3. The figure shows a graph where the x variable is speed in 5 mph increments (0-
80), the y variable is acceleration in 0.5 mph/sec (+10 to -10) increments, and the z
variable is the fraction of activity.
7.0%
30 Velocity (mph)
15
Acceleration (mph/sec)
Figure 3.3 - Speed / Acceleration Profile, Interstate Ramp, LOS D
3.4.3.4. Road Grade
The impacts of road grade on emissions are included in the model design.
Road grade affects vehicle emissions by impacting the load on the engine. Gravity
exerts a force on a vehicle that must be counteracted to maintain a constant speed.
Road grade is not included in mandated emission models because tests on the actual
effects have not been completed and because metropolitan areas do not maintain
spatially defined road grade estimates. Although grade impacts on emission rates are
44
-------
being researched, results are not available at this time. However, the effects of
acceleration on emissions have been quantified. Therefore, the secondary effects of
grade on acceleration can be included in the conceptual design.
The effects of grade on acceleration can be quantified by the equation:
Acceleration (mph/sec) = 22.15 (mph/Sec) x (Gradient (road))
where 22.15 (mph/sec) represents acceleration due to gravity. For example, a
vehicle wishing to maintain a constant speed along a 5% road grade must accelerate
1-11 (mph/sec) to counteract deceleration due to gravity.
Road grade data, while not currently comprehensively available for urban
areas, is information that can be collected using global positioning systems (GPS)
[Awuah-Baffour, 1997]. Given the expected importance of grade in affecting running
exhaust emission rates, it is likely that the new GPS strategies will be employed by
metropolitan areas in the next few years.
Including vehicle activity impacts resulting from road grade, even if not fully
developed, provides an important step in emission model development. Strategies that
are used for developing connections between road grade data and other road
characteristics will act as guides in the development of future load-based models.
3.4.3.5. Temporal Variability
The temporal variability in estimates of vehicle activity is highly inaccurate
because it relies on traditional travel demand forecasting models [see section 2.3.1].
Travel demand forecasting models are designed and operated to predict peak hour
travel or daily travel. These are the primary temporal aggregation levels used by
transportation planners and traffic engineers. The Travel Model Improvement
Program (TMIP) administered by the US Department of Transportation is researching
strategies for travel models to better predict activity during off-peak hours.
The ability of the research model to predict hourly emissions will rely heavily
on accurate vehicle activity measures throughout the day. Until progress is made in
the TMIP research, this emissions model will be unable to accurately incorporate off-
peak travel. However, an intermediate step between existing and future models is
possible. Many MPOs have developed estimates of hourly or subhourly travel demand
factors based on travel survey data. These regional factors by trip type can be used to
disaggregate daily trip generation into hourly intervals. Data on the variability of road
volume throughout the day are available from departments of transportation for many
major roads. Although average speed cannot be predicted to determine LOS F during
45
-------
The 'conceptual correctness' of the vehicle activity refers to the accurate
portrayal of the actions of an aggregate group of vehicles, or, in other words, the
ability to predict the distribution of activity in a zone or on a link. The largest source
of error comes from the travel demand forecasting model. Heavy reliance on the
model transfers errors in trip generation estimates, road volumes, and road speeds to •
the emission models. The research approach attempts to lessen the impact of these
errors by using modal activity measures when possible, and by disaggregating trip
origins to appropriate land uses. The use of temporal factors causes substantial error
in the activity estimates because the factors are used region-wide and lack spatial
variability. While better than current practice, the approach results in significant
problems with accurate representation of emission-specific vehicle activity.
• Vehicle Activity Conceptual Completeness
The 'conceptual completeness* of the representation of emission-specific
vehicle activity refers to the ability of the modeling approach to capture all of the
important activities. The largest gap in the completeness of the representation occurs
on non-highway or ramp roads; Few speed and acceleration profiles are available for
major and minor arterials. An enhancement to the travel model is a linkage between
trip origins and the major road network. Minor road shortest paths allow vehicle
activity to be estimated between the zonal-based starts and the lineal-based running
exhaust. Further, inadequate representation of activity around signalized and
unsignalized intersections may cause the exclusion of a large source of emission-
specific vehicle activity. Until other data are available that can help determine these
operations, the completeness of the vehicle activity estimates will be poor.
• Vehicle Activity Enterprise Awareness
The estimates of vehicle activity can be tied to other enterprises through the
zonal aggregations and road network. The road network maintains a variety of
locational parameters including street address and travel model identifiers.
3.4.4. Facility Emissions
Facilities are divided into zones and lines corresponding to the previously
mentioned emission modes of engine starts and running exhaust (respectively).
Facility estimates are used to allocate emission production to those vector spatial data
structures currently used by transportation planners. By tying emission production
estimates to facilities, tasks regarding research, reporting, validation, or control
strategy development are made easier.
46
-------
3.4.4.1. Engine Start Zonal Facility Estimates
Zonal facilities include the zonal representations of TAZs, land use, and
Census blocks. The model design allows for other zonal designations to be included,
but only the three mentioned have been required. The zones have been included in the
definition of facilities because they are used by planners to aggregate socioeconomic
information. While running exhaust emissions occur within zones, they are better tied
to modal activity that occurs on the road. Engine starts, however, occur at trip origins,
generally characterized with point or zonal information.
Figure 3.4 schematically represents the portion of a vehicle's emission profile
represented by zonal facilities. The exhaust engine start estimates are modeled as a
'puff ( all start emissions allocated to the trip origin). While start emissions are
actually dispersed through the network as a vehicle travels, research has not identified
a strategy for correct spatial allocation. However, the role of this model is the study of
emission production by automobiles, not air quality. It may be more useful for
planners and/or researchers to have start emissions tied to the point of origin, thus
allowing linkages to other zonal information.
Engine start emission rates are included in the research model based on results
from the statistical model [see section 2.2.3]. Emissions in grams per start are
estimated using the regression tree mentioned in sections 3.4.2.3 and 3.4.2.4. Six
technology group trees exist for engine start emissions, all based on vehicle technology
characteristics. Each emission estimate has established confidence bounds that can be
translated back through the model to assess accuracy. The technology characteristics
used in the tree process are listed below:
47
-------
MY = Model Year
EMM - Emission Control Equipment, I-none, 2-oxi, 3-cat, 4-oxi&cat
FINJ = Fuel injection equipment, 1-port, 2-carb, 3-throt
CID = Engine Size, Cubic Inch Displacement
TWT = CERT test weight, Ibs.
Engine Start
Acceleration
Enrichment
Grade
Enrichment
Engine
Rate
Engine On
Engine Off
Time (seconds)
Figure 3.4 - Engine Start Emission Portion
The resulting technology groups are mutually exclusive and listed below:
• CO Normal:
1. MY< 1981, TWT < 3250
2. MY < 1980, TWT >= 3250, TWT < 4375
3. MY < 1980, TWT >= 4375, CID < 351
4. MY < 1980, TWT < 4375, CID >= 351
5. MY>= 1980, TWT >= 3250
6. MY = 1981, TWT >= 3688, CK)< 131
7. MY= 1981, TWT < 2938, CID < 131
8, MY =1981, TWT >= 2938, TWT < 368° CID < 131
9. MY >= 1982, MY < 1987, TWT < 368!
10. MY >= 1982, MY < 1987, TWT >= 36i>-,
ll.MY>=1987
48
-------
• CO High:
1. CID < 116, FINJ < 2
2. CID<116,FINJ>=2
3. CID>=116,CID<134
4. CID>=134,CID<258,FINJ<2,MY<1986
5. CID >= 134, CID < 258, FINJ = 2, TWT < 3563, MY < 1986
6. CID >= 134, CID < 258, FINJ = 2, TWT >= 3563, MY < 1986
7. CID >= 134, CID < 258, FINJ = 2, MY >= 1986
8. CID >= 134, CID < 258, FINJ >= 3
9. CID >= 134, CID >= 258
• HC Normal:
1. MY < 1980, TWT < 4125, CID < 154
2. MY < 1980, TWT < 4125, CID >= 154, CID < 241
3. MY < 1980, TWT >^ 4125, CID >= 241
4. MY < 1978, TWT >=4125
5. MY >= 1978, MY < 1980, TWT >= 4125
6. MY >= 1980, EMM < 4, CID < 171
7. MY>=1980,EMM<4, CID >= 171
8. MY>= 1980, MY < 1988, EMM >= 4, CID < 98
9. MY >= 1980, MY < 1988, EMM >= 4, CID >= 98, CJ£> < 102
10. MY >= 1980, MY < 1988, EMM >= 4, CID >= 102
ll.MY>=1988
• HCHigh:
L MY < 1980
2. MY>= 1980, FINJ < 3, CID < 196
3. MY >= 1980, FINJ < 3, CID >= 196, CID < 258, MY < 1987
4. MY >= 1980, FINJ < 3, CID >= 196, CID < 258, MY >=1987
5. MY >= 1980, FINJ < 3, CID >= 258, MY < 1983
6. MY >= 1980, FINJ < 3, CID >= 258, MY >= 1983
7. MY>= 1980, FINJ >= 3, TWT < 2688
8. MY >= 1980, FINJ >= 3, TWT >= 2688, MY < 1988, CID < 192, TWT < 3063
9. MY >= 1980, FINJ >= 3, TWT >= 3063, MY < 1988, CID < 192
10. MY >= 1980, FINJ >= 3, TWT >='2688, MY < 1988, CID >= 192
ll.MY>= 1980, FINJ >= 3
• NOx Normal:
1. EMM<3
2. EMM >= 3, EMM < 4, CJD < 230
3. EMM >= 3, EMM < 4, CID >= 230, CID < 245, FINJ < 2
4. EMM >= 3, EMM < 4, CID >= 230, CID < 245, FINJ >= 2
49
-------
5. • EMM >= 3, EMM < 4, CID >= 245
6. EMM >= 4, CID < 122
7. EMM >= 4, CID >= 122, CID < 138
8. EMM >= 4, CID >= 138, CID < 146
9. EMM >= 4, CID >= 146, CID < 152
10. EMM >= 4, CID >= 152, CID < 213
11. EMM >= 4, CID >= 213, CID < 288
12. EMM >= 4, CID >= 288
• NOxHigh:
1. EMM < 3, CID < 334, MY < 1980
2. EMM < 3, CID < 334, MY >= 1980
3. EMM < 3, CID > 334
4. EMM >= 3, CID < 137, MY < 1987
5. EMM>=3, CID< 137,MY>= 1987
6. EMM >= 3, CID >= 137, CID < 152
7. EMM >= 3, CID >= 152, CID < 230
8. EMM >= 3, CID >= 230, CID < 232
9. EMM >= 3, CID >= 232
Zonal estimates of fleet characteristics are divided into the previous technology
groups. Each technology group fraction is multiplied by the number of trip origins that
occur in the zone. The resulting number of trip origins by technology group is
multiplied by the associated gram per start emission rate. The resulting emissions of
CO, HC, and NOx are reported for a typical weekday (Tuesday - Thursday) on an
hourly basis. The typical weekday limitation is a result of the travel demand modeling
process, as few models predict weekend or Friday travel.
3.4.4.2. Minor Road Zonal Facility Estimates
Minor road zones [see section 3.4.1.2] are used to spatially represent the
portion of running exhaust emissions that occur between the trip origin and the roads
modeled by the travel demand forecasting network. Available minor road vehicle
activity information is limited because it is not explicitly modeled in the travel
forecasting process, and there are no existing measures of modal activity available for
local roads. Lower traffic densities and lower average speeds suggest that the actual
portion of running exhaust en. dons occurring on local roads may be small.
However, little evidence is avai. ,e to draw conclusions about the impacts of local
road driving. This limitation fc^es a scaled back version of local road emissions
modeling.
50
-------
3.4.4.3. Lineal Facility Estimates
Lineal facilities are roads that are modeled in the travel demand forecasting
model. On-road fleet distributions and predicted traffic flow parameters are used to
generate road segment specific estimates of CO, HC, and NOx. Figure 3.5 shows the
portion of the emission spectrum represented by linear features. Minor road running
exhaust emissions and major road running exhaust emissions estimate the same
pollutant and, combined, predict total running exhaust, emissions. Network
characteristics determine the amount of the running exhaust portion that is allocated to
minor zones.
Engine Start
Acceleration
Enrichment
Engine
Rate
Grade
Enrichment
Hot Stabilized
and Running Losses
Engine On
Engine Off
Time (seconds)
Figure 3.5 - Running Exhaust Emission Portion
Emission rates for running exhaust come from two sources, the statistical
approach used for start emissions, and the SCF approach, used by currently mandated
models. The purpose for including both approaches is to allow user flexibility and to
provide a platform for comparison. Vehicle activity measures for both emission rate
approaches come from the same source, although different variables are needed. The
SCF approach needs average speed, while the statistical approach uses a variety of
51
-------
other modal parameters. The same vehicles are aggregated and used for the estimated
fleet distribution, although different technology group definitions exist.
The technology groups (see section 3.4.4.2 for variable definitions) for the
statistical model were determined similarly to those described in the engine start
section. The definitions are as follows:
• CO Normal:
1. EMM < 4, MY < 1979
2. EMM < 4, MY >= 1979
3. EMM>=4, CID< 146, MY < 1979
4. EMM>=4,CID<146,MY>=1979
5. EMM >= 4, CID >= 146, MY < 1979
6. EMM >= 4, CID >= 146, MY >= 1979, MY < 1985
7. EMM >= 4, CID >= 146, MY >= 1985, MY < 1987
8. EMM >= 4, CID >= 146, MY >= 1987
• CO High
1. EMM < 4, FINJ < 3, TWT < 3375
2. EMM < 4, FINJ < 3, TWT >= 3375
3. EMM < 4, FINJ >= 3
4. EMM >= 4, TWT < 3313
5. EMM>=4, TWT >= 3313
• HC Normal:
1. MY < 1985
2. MY>= 1985, MY < 1987, TWT < 3188
3. MY >= 1985, MY < 1987, TWT >= 3188
4. MY>= 1987, MY < 1990, CID < 143
5. MY >= 1987, MY < 1990, CID >= 143, CID < 196
6. MY >= 1987, MY < 1990, CID >= 196
7. MY >= 1990, TWT < 3563, CID < 143
8. MY >= 1990, TWT < 3563, CID >= 143, CID < 186
9. MY >= 1990, TWT >= 3563, CID < 143
10. MY >= 1990, TWT >= 3563, CID >= 143, CID < 186
11. MY >= 1990, CID >= 186, CID < 196
12. MY>= 1990, CID >= 196
• HCHigh:
1. MY < 1984, EMM < 4
2. MY>= 1984, MY < 1985, EMM < 4
3. MY>= 1985, MY < 1988, EMM < 4
52
-------
4.' MY < 1985, EMM >= 4, TWT < 4000
5. MY < 1985, EMM >= 4, TWT >= 4000
6. MY >= 1985, MY < 1988, EMM >= 4
7. MY >= 1988
• NOx Normal:
1. MY < 1989, CID < 154
2. MY >= 1989, MY < 1994, CID < 154
3. MY >= 1994, MY < 1995, CID < 154
4. MY >= 1995, CID < 154
5. MY < 1994, CID >= 154
6. MY >= 1994, MY < 1995, CID >= 154
7. MY >= 1995, CID >= 154
Each engine start technology group will have a gram per start emission rate.
Running exhaust emissions rates depend on modal operation. The regression tree used
to determine running exhaust emissions rates includes modal parameters. The modal
parameters are:
• AVGSPD The average speeding miles per hour.
• PKE>X The fraction of activity with positive kinetic energy (speed*acceleration)
greater than X mph2/sec.
• POW>X The fraction of activity with power (speed * acceleration) greater than X
mph3/sec.
• ACC>X The fraction of activity with acceleration greater than X mph/sec.
• DEC>X The fraction of activity with deceleration greater than X mph/sec.
• CRZ>X The fraction of activity with zero acceleration and speed greater than x
mph.
• IDLE The fraction of the activity with zero acceleration and zero speed.
For each road segment and each hour, modal variables are determined. Road
segment-specific technology groups and modal variables are combined to develop the
fraction of activity with specific emission rates (grams per second). Total hourly
travel time is calculated and segmented by the fraction of the vehicles with each
emission rate.
The SCF emission rate approach uses a 'look-up' table created from running
MOBILESa for a series of model year distributions and average speed distributions.
The vehicle characteristics developed for the model include model year as a variable,
allowing the creation of model year technology groups. The vehicle activity
component of the model allows the estimate of average speed to be available for every
53
-------
road segment and every hour. By running MOBILESa for each combination of
average speed and model year distribution, gram per second emission rates will be
developed.
• Facility Emissions Estimates Conceptual Correctness
Conceptual correctness for facility emission estimates is the ability of the
approach to accurately estimate actual emission production. Sources of inaccuracies
come from three places; the quality of the input data, the model's manipulation of the
data, and the errors in the development of emission rates. The quality of the input data
and the ability of the model to manipulate them into a usable form result in large
errors. All of the problems mentioned in the previous sections culminate in substantial
error. Errors in the development of the emission rates are the result of
incomprehensive test datasets and unrepresentative operating profiles. Emission rates
have confidence bounds associated with them, allowing some measure of the
variability of the errors. As in the representation of vehicle activity, the conceptual
correctness of the faculty emission rates is defined by the aggregate level of the
estimate. While the accurate assessment of a single vehicle may be poor, the aggregate
estimate may prove better. The SCF emission rates have been strongly criticized as
being insensitive to important activity. Thus, the SCF emission rates have less
'correctness' than the statistically based approach that includes modal parameters.
• Facility Emissions Estimates Conceptual Completeness
The 'conceptual completeness' of the exhaust emissions estimates is fairly
strong. All automobiles will fit into engine start, running exhaust, and SCF
technology groups. All operating vehicles will fall in the speed and acceleration
profile identified for the particular conditions. The only source of incomplete
representation of emission rates is due to problems that have already been mentioned
regarding, the emission test dataset and the cycle test design. All represented vehicles
can be assigned an emission rate.
• Facility Emissions Estimates Syntactic Correctness and Completeness
The communication of the facility emission estimation component relies on
clear definitions and visual representation. The terminology used to describe the
technology groups and other input parameters can be clearly communicated if the user
is given concise explanations of all the parameters. The spatial component of the
facility estimates can be clearly communicated using GIS and by using the input data
spatial structures.
• Facility Emissions Estimates Enterprise Awareness
The estimates have strong 'enterprise awareness' due to the ability of the
estimates to be aggregated to any of the original input spatial structures (TAZs,
Census, etc.). If needed, the estimates can also be presented as emissions per unit of
54
-------
distance or area to aid in translating the other locational parameters. This step is
improved through the translation to raster structure in the next section; however, the
base information needed is available at this level.
3.4.5. Emissions Inventory
The role of the emissions inventory module is to prepare the facility-based
emission estimates for input into gridded photochemical models. An important
component of the entire modeling process is the ability to aggregate estimates to a
user-defined grid cell size. The most efficient technique for accomplishing this task is
to convert the engine start, minor road running exhaust, major road running exhaust,
and major road SCF emission estimates to raster data structures. Once in a raster
structure, . developing gridded estimates for inventory reporting is fairly easy.
Conversion of the data from vector to raster is a tool available in many of the larger
GIS software tools. After conversion, total mobile source emission estimates are
calculated for the entire area. Engine starts, minor road running exhaust, and major
road running exhaust emissions are used to develop totals.
The tools available in the GIS for conversion make some assumptions about
the vector data that may not be desired. Problems occur with direct conversion,
especially for linear structure. Straight conversion is possible, but grid cells take on
the value of the largest feature, or largest portion within its boundaries. For a linear
feature, this means that all cells that represent the line will have the same emission
value or rate. However, the line can bisect the cell at any point, resulting in variations
in the cell's ability to properly identify the portion of the road that falls within its
boundaries. Similar issues occur with polygons along their boundaries. The smaller
the grid cell, the lesser the problem. However, gridded inventories require grid sizes
as large as 4-5 square kilometers, much larger than the anticipated zonal and lineal
structures.
One way around the problem is to intersect the zonal and lineal structures with
a vector grid. Once intersected, all emission values falling within the grid cell
boundaries can be weighted by area or length and summed. The resulting vector grid
data are then converted to raster cells with cell sizes equivalent to the vector grid size.
The final raster datasets are individual 'layers' of each pollutant by hour and
emission 'mode' (totals, engine starts, etc.). Tools available in the GIS allow for the
development of special visualization interfaces that can create and query two and three
dimensional images of the various databases.
55
-------
• Gridded Emissions Conceptual Correctness
For the final module, 'conceptual correctness' refers to the ability to maintain
data integrity while converting from raster to vector. The deterministic, vector
approach to developing gridded estimates ensures that few errors are introduced during
the process. There are also problems with grid cells that fall on the boundaries of the
study area. Cells that overlap study area boundaries will only have values for those
emissions in the study area, causing the cell value to be underestimated. As long as
cells lie completely within the boundaries, this is not a problem. The accuracy of the
gridded estimates is a function of all of the previously discussed problems.
• Gridded Emissions Conceptual Completeness
The completeness of the process is Only limited when grid structures do not
encompass or lie within the entire study area.
• Gridded Emissions Enterprise Awareness
Aggregating and rasterizing the emission estimates allow the storage and
communication of specific emission production intensities. Previously, a zone would
represent only emission produced by 'zonal' information like trip generation estimates.
With the gridded emissions, every location has an estimate of the total emissions as
well as the specific modes. The flexible locational parameters allow the estimates to
be translated to other areas.
3.5. Conclusion
This chapter defined the parameters and quality assurance measures for the
design of a comprehensive modal emissions model. Initially, a detailed list of issues
to be addressed by the model was compiled from background research. This list of
issues led to a model design that incorporated those parameters while maintaining a
flexible, comprehensive and accurate account of the science being modeled. In
developing the design, GIS became a powerful tool for data preparation, storage, and
analysis.
56
-------
4. MODEL DEVELOPMENT
This chapter discusses the design of the research model, hereafter referred to as
the Mobile Emission Assessment System for Urban and Regional Evaluation
(MEASURE). The chapter provides descriptions of the data, processes, algorithms,
and files. The chapter will review each of the input files, pre-processing steps,
program modules, and program code. By the end of the chapter, the reader should
have a clear vision of the model scope and process.
The purpose of the MEASURE is to. provide researchers with a tool for
measuring the air quality impacts of urban and regional transportation policy and
development changes. This model is not intended to be used directly for conformity or
inventory reporting, but for use in a research environment by scientists exploring the
transportation and air quality relationship.
MEASURE will produce hourly transportation facility-level and gridded
automobile exhaust emission estimates of carbon monoxide (CO), hydrocarbons (HC),
and oxides of nitrogen (NOx). The model develops these estimates based on a
geographic area's vehicle registration data, accurate digital road dataset, TRANPLAN-
(travel demand forecasting) model output, and zone-based socioeconomic data (i.e.,
US Census Blocks, Land Use). The outputs of the model include dbase data files and
gridded inventories in ARC/INFO raster 'grids'. ARC/INFO software features allow
the user to develop maps and other visualization tools.
MEASURE currently includes:
• Modal (cruise, acceleration, deceleration, idle) activity
• Estimates of CO, HC, and NOx
• -100 emission specific technology groups
• Separate tree-based regression engine start and running exhaust emission rates
• Comparative MOBILESa SCF running exhaust estimates
• Impacts of grade on acceleration
MEASURE currently does not include:
• Non-automobiles
• PM10orPM2.5
• Evaporative emission estimates
• ' Vehicle deterioration effects
• ' Impacts of grade on engine load
57
-------
• Impacts of accessory load
• Intersection specific estimates
Included, but technically weak components:
• Hourly, on-road traffic volume estimates
• Hourly, on-road traffic flow estimates
Running and managing, the model require technical skills in geographic
information systems (ARC/INFO 7.0), transportation planning practice and software
(TRANPLAN), air quality modeling (MOBILE 5a), C programming, and UNIX and
MS-DOS operating systems. Substantial data collection and processing activity is
required prior to model operation and is discussed in detail in the next section.
MEASURE is divided into 12 modules (see Figure 4.1). The modules and
their associated input and output data files are managed by a single ASCII description
file called 'Makefile'. The 'Makefile' is executed by a UNIX system utility called
make (discussed in detail later). The modules roughly follow a tree structure in that
one module may depend on the outputs of other modules. The modular structure
allows individual components to be executed independently or in a fully connected
process. The modules are grouped into five tiers: the spatial environment, the fleet
characteristics, the vehicle activity, the facility emissions, and the emission inventory.
58
-------
GIS-Rasp.d Model nf A utnmohile Exhaust Emissions
Input Data:
Spatial
Environm ent
Fleet
Characteristics
Vehicle
Activity
Facility
Em ission
Em ission
Inventory
• Census Block Data • Road Network Data
• Land Use Data • Travel Demand Forecasting Network
• Traffic Analysis Zone Data • Speed /Acceleration Profiles
• Vehicle Registration Data • Road Grade Data
1
3
5
8
Zonal Module
2
• r
K
«
Zonal TG Module
4
• c
t
Engine Start
Activity Module
o 7
_A Minor Road A Ru
^ Activity Module ^ Ac
| ^ 10a '
Engine Start
Emissions Module
y
Minor Road Agg. Moda!
Running Exhaust Running Exha
Module Module
1
11
T
Hourly Gridded Starts, Hot
Stabilized, Enrichment, and
Total Emissions Module
oad Module
©
)n-road TG
Module
^
ining Exhaust
tivity Module
"^ ^ 10b
JS, SCF Running
Exhaust Module
_f
Figure 4.1 - Model Design
59
-------
If all input files and associated software are in place, the modules are executed
in the appropriate order by typing 'make emissions' at the UNIX command line. The
test dataset (100 sq. km in northeast Atlanta) should take approximately four hours to
complete on a Sun SPARCstation 10. A larger urban area may take as long as 24 to 36
hours to complete a full execution.
4.1. Input Files
This section will describe the datasets, fields, and directories needed to be
resident in the system. A substantial amount of data collection and processing effort
needs to be conducted prior to model operation. Several tools and guidelines are
included to aid this process. As a general note, all ARC/INFO coverages and files
containing coordinates should use the same geographic projection and use meters as
the base unit.
4.1.1. Directory structure
The directory structure consists of one home directory and several data
directories. The data directories are:
• zone: stores all zonal data and coverages
• road: stores all lineal data and coverages
• grid: stores all vector grid data
• em: stores all emission estimate data
• tg: stores all technology group data
• gra'de: stores all road grade related data
• raster: stores all raster data
• raw: place to store backup copies of data and programs
• ami: stores all AML code
• code: stores all C code
• templates: stores a number of INFO file templates
• sa: stores all speed / acceleration profiles
• modalmats: stores all modal matrices
• lookup: stores all ASCII lookup files
• temp: stores temporary files used during program runs
60
-------
4.1.2. Zone.twt and zip.twt
Two ASCII files, zone.twt, and zip.twt, need to be created and stored in the tg
directory. These files represent the area's vehicle characteristics. Zone.twt is a list of
successfully addressed geocoded vehicles and an ID number representing the
registered zonal location. Zip.twt are those vehicles that were not matched and
therefore only the registered ZIP code location is included. The ASCII files have the
same comma-delimited structure ,of:
• zone or ZIPcode
* vehicle identification number (VIN)
• model year
• emission control code
• fuel delivery system code
• engine size (cubic inch displacement)
• vehicle dynamometer test weight
• CO high emitter flag
• HC high emitter flag
• NO high emitter flag
Creating these files from an area's vehicle registration dataset is a lengthy, one-
time process and therefore removed from the formal model. The first step in the
process is to acquire the area's registration datafile that contains VIN, street address,.
and ZIP code. For a large urban area, this initial file can consist of millions of records
(2.2 million for Atlanta). The file needs to be address geocoded (with offset) using
tools available in ARC/INFO or other software, and an accurate, up-to-date, road
database. Successful matches are stored in an ARC/INFO point coverage with the
VIN. Unsuccessful matches are stored in an ASCII file of ZIP code and VIN. Keep
track of the match rate, as it will be important later in the model. The point coverage
is then overlaid with any zone coverage, preferably of census blocks, using
ARC/INFO's identity command. The resulting zone-id and VIN are written to an
ASCII file. The two ASCII files are processed through a series of programs written by
individuals at Georgia Tech. These programs read these files, decode the VINs using
vendor software, flag a random sample as high emitter for the three pollutants, and
write the results to ASCII files called zone.hne and zip.hne. A separate program,
hne2t\vt.c, reads these files and uses a lookup table to add vehicle dynamometer test
weight.
4.1.3. Allroads (ARC/INFO coverage)
The allroads coverage is a line database of all roads in the area and should be
as accurate as possible. It should be stored in the road directory. The database
contains, in addition to standard ARC/INFO fields, the key fields called arid and tfid.
61
-------
These fields are identifier fields and rarep used Jo Jink datasets back to the individual
road segment locations. Arid is unique for each road segment. Tfid identifies the
corresponding travel demand model link. Because travel demand models usually
model major roads only, not all lines in allroads will have a non-zero tfid. Further,
some travel model links will span a number of lines in allroads resulting in lines that
have the same tfid. Establishing the tfid on the lines is completed through a manual
process called conflation. The process of conflation involves selecting each travel
model link, selecting all corresponding road segments, and joining the ids.
4.1.4. Tdfn.dat (INFO file)
The tdfn.dat INFO data file is a table stored in the road directory with the
following items:
• tfid - the key field to link to allroads
• assign_gr - the road classification code (I-interstates, 2-ramps, 3-major arteries,
4-minor arteries)
• abvolume - the oneway 24 hour volume
• abspeed - the oneway 24 hour average speed
• abcap - the oneway road capacity
• bavolume - the otherway 24 hour volume (blank if divided road)
• baspeed - the otherway 24 hour average speed
• bacap - the otherway road capacity
This file is created by converting the area's travel demand forecasting model
outputs into an INFO format. Programs by Georgia Tech are available that can
complete this process for TRANPLAN models. If other travel modeling software is
used, one may have to contact the vendor for a conversion package, or develop
customized software that creates datafiles compatible with ARC/INFO's 'generate'
and 'add from' commands.
4.1.5. Census (ARC/INFO coverage)
The census polygon coverage (stored in the zone directory ) can be any zonal
structure that contains a 'household' field. US Census blocks are preferred because
they are available around the country and provide substantial detail with regard to
population and household information. The census coverage contains the following
fields:
• cbid - the key field to link associated databases together
• hu90 - the household field
Census data can be acquired from the US Census Bureau.
62
-------
4.1.6. Land use (ARC/INFO coverage)
The landuse polygon coverage (stored in the zone directory) is a database of
residential, non-residential, and commercial land uses. The coverage needs to contain
only a single attribute field of lu. Lu is a character field that consists of 'RES' if the
zone is residential, 'COM' if the zone is commercial, and 'NONRES' if the zone is
non-residential. Areas without identifiers will be considered non-residential.
4.1.7. ZIP code (ARC/INFO coverage)
The Zipcode polygon coverage (stored in the zone directory) is a database of
postal ZIP codes. The coverage needs only a single field called Zipcode and should be
populated with current five-digit ZIP codes.
4.1.8. TAZ (ARC/INFO coverage)
The TAZ polygon coverage (stored in the zone directory) is a database of traffic
analysis zones used in the travel demand forecasting process. The coverage needs a
single key field called tzid and should be unique for every zone. Developing this
datafile will probably require assistance from the local planning organization.
4.1.9. TAZ.dat (INFO file)
The TAZ.dat file stores all the trip generation information from the four step
travel demand model. The fields are as follows:
• tzid the key field used to link the data to other zones
• hbwprd home-based work productions
• hbshprd home-based shopping productions
• hbgsprd - home-based grade school productions
• hbuprd - home-based university productions
• nhbprd - non-home-based productions
• hbwatt home-based work attractions
• hbshatt - home-based shopping attractions
• hbgsatt home-based grade school attractions
• hbuatt home-based university attractions
• nhbatt - non-home-based attractions
The productions and attractions fields listed above represent standard travel
demand forecasting terminology for trip type. The fields should represent the 24-hour
trip type quantities by TAZ. This file, like the TAZ coverage, needs to be converted to
INFO from whatever format the planning organization provides.
63
-------
4.1.10. Landmarks(ARC/INFO coverage)
The landmarks coverage is a point database of educational institutions. The
points should have an identifier field that shows whether or not it is a university or
grade school. The determination of which schools to include depends on the definition
of the trip types from the travel demand forecasting model. They are used primarily to
spatially allocate home-based-university trips and home-based-grade school trips.
4.1 .1 1 .- Srid (ARC/INFO coverage)
The -&$d coverage will represent the spatial structure of the inventory
estimates. Grid cells of any size can be used; however, they should not be smaller
than the published accuracy of any of the input coverages. The coverage should
contain one attribute called gdid. Gdid will be used as a key item to aggregate
estimates from the zones and lines.
4.1.12. Grade.xyandgrade.gr
The ASCII files grade.xy and grade.gr store road grade information collected
from GPS units. The grade.xy file contains comma delimited fields of grade-id,
x_coordinate, and y_coordinate. The coordinates must match the geographic
projection system and units used by the coverages. The grade.gr file contains comma
delimited fields of grade-id and grade. The files must be separate in order to read
them into ARC/INFO as a point coverage. The model joins them in the mr-act.aml
process.
4.1.13. Lookup files
Three lookup files are provided, temporal/actors (INFO file), scf.csv (ASCII
file), and twt.lu (ASCII file). These files can be used in any model run, but may be
updated. Temporal. factors provides the breakdown of vehicle volumes and trips by
hour. Scf.csv provides MOBILE 5a speed-corrected emission factors by 5 mph speed
increment and model year. It is used in developing the SCF running exhaust emission
estimates. Twtlu.asc is a file of vehicle make, model, model year, and test weight. It
is used outside MEASURE to add the test weight field to the vehicle characteristics
files.
64
-------
4.2. The Makefile
The Makefile is the most important file in the system to become familiar with.
It manages the entire modeling process by checking file and code dependencies. The
Makefile is interpreted by a system utility in-UNIX operating system software (make).
Make identifies file dates and times to determine when updates have been made, and,
if needed, calls a series of actions. For example, the first dependency relationship in
the Makefile is listed as follows:
$(em)/grid-em.dbf : $(em)/scf-em.dbf \
$(em)/sz-em.dbf \
$(em)/mr-em.dbf \
$(em)/mz-em.dbf \
$(ami)/grid-em.ami
/bin/rm -r $ (raster)
/bin/mkdir $ (raster)
/bin/rm -r $(temp)
/bin/mkdir $(temp)
$(arc) "&r $(ami)/grid-em.ami"
Make first checks to see if a file 'grid-em.dbf exists, and if it does, it compares
it with the last update of the other files. If one or more of the other files has a newer
date than 'grid-em.dbf, the last five lines are executed. If 'grid-em.dbf is current
with respect to the other files, the code is not executed. The value of this is that it
saves time in large complicated multi-program processes because only those portions
that have been updated are executed.
The Makefile is segmented into- twelve parts that represent the twelve modules
listed in Figure 4.1. The twelve parts are called:
• gridded_emissions
• scf_emissions
• eng_starts_emissions
• maj_rds_run_exh_emissions
• min_rds_run_exh_emissions
• min_rds_activity
• eng_starts_activity
• maj_rds_activity
• run_exh_tech_groups
• eng_starts_tech_groups
• rds_spatial_environment
• zonal _spatial_environment
65
-------
Any of the above parts can be executed by typing 'make ' and that
particular module, and any non-up-to-date module it depends on, will be executed. By
typing 'make gridded_emissions' all the parts are checked and executed if needed
because the final gridded emission estimates depend on all of the other components.
In addition, two other parts have been added to aid in modeling. 'Make clean'
will remove all data files except the original input files. This can be handy when one
wishes to start with a clean slate. 'Make programs' will compile all the 'C' programs.
This is handy when editing code and you want to make sure it works before running
the
Prior to running the Model on a new system, edits to the Makefile must be
made. The first section entitled 'Variable Definitions' is as follows:
dir = /proteus/home/wbachman/gismodel2
tg = $(dir)/tg
zone = $( dir) /zone
road = $ (dir) /road
1m = $ (dir) /landmarks
em = $ (dir) /em
grid = $ (dir) /grid
c = $ (dir) /code
ami = $ (dir) /ami
raster $ (dir) /raster
temp = $ (dir) /temp
arc = /miranda/ceesri/arcexe70/programs/arc
cc = /opt/SUNWspro/bin/cc
The variable 'dir' in the first line must be updated to the current path where the
Makefile and other directories reside. Also, the 'arc' variable must be updated to the
correct executable path for ARC/INFO. The same update must be made for the 'cc'
variable, identifying the location of the C compiler.
4.3. The Modules
The modules, although separate, are linked together through dependent and co-
dependent files and programs. Even though they are discussed separately, and
function independently, their design and operation is affected by inputs and outputs of
other modules.
66
-------
4.3.1. Zonal Environment Module
The zonal environment module's purpose is to establish a polygon database
called sz that will be used to develop engine start emission estimates. Figure 4.2
represents the entities involved in MEASURE. Figure 4.3 represents the flow of the
code. The polygon database is created by spatially joining four input coverages,
census, taz, landuse, and ZIPcode. The process of joining the databases involves the
use of the ARC/INFO command 'identity'. The input coverages may have relative
inaccuracies that can cause 'sliver' polygons to be created. Operationally, this is not
important, but it does affect relative accuracy. A line on one coverage may represent
the same feature as the line on another coverage; however, the spatial representations
of the line may differ. To lessen the impact of this problem, a 'fuzzy' tolerance is
designated. Any lines that fall within the tolerance distance will be considered the
same feature. The fuzzy tolerance should be set to the size of the worst reported
absolute accuracy. In most instances, this will be the Census Bureau files that
maintain an accuracy of 30-100 meters.
The module also creates a data file that maintains various attributes
accumulated through the integration of the four input databases. These fields are land
use (RES, NON-RES, and COM), housing unit per square kilometer, and all the TAZ
trip generation estimates. The housing unit rate is multiplied by the sz area to predict
the number of households. The amount of error introduced by disaggregating
household data is a function of the size and accuracy of the original household
database. If the Census blocks are used, a common source of large scale household
data, the disaggregation errors will be small because the new zones will be similar to
the blocks.
4.3.2. Road Environment Module
The purpose of the road environment module is to create the spatial structure
used to represent running exhaust emissions. The road environment module divides
the allroads coverage into roads modeled by the area's travel demand forecasting
model, and those that are not modeled. The modeled road segments are used to create
the mr (major road) line coverage. The mr lines form the boundaries for mz (minor
zone) polygons that represent a zonal aggregation of minor roads. Figures 4.4 and 4.5
represent the entities and program code flow, respectively.
The module does not result in any loss of accuracy. Some minor zone
polygons become insignificant due to the techniques used for generation. Median
areas bounded by two parallel roads become minor zones. They bias the network
travel distances slightly (determined later).
67
-------
4,3.3. Zonal Technology Groups Module
The purpose of the zonal technology group module is to convert the outputs of
the vehicle characteristics process into zonal technology groups. The vehicle
characteristics process and zonal TG modules are displayed in Figures 4.6 and 4.7.
Figures 4.8 and 4.9 describe the pre-processing steps. The module reads the two
ASCII input files ofzone.twt and zip.twt and creates four dbase files; sztg.dbf, retg.dbf,
scftg.dbf, and regiontg.dbf. The module contains four programs; techgr.c, regiontg.c,
jointg.c, and sz-tg.aml. Techgr.c (Figure 4.10) is executed first and it assigns each
vehicle into a technology group for each pollutant and summarizes by zone or ZIP
code. The resulting distributions are written to a series of ASCII files. The regiontg.c
program is similar in structure to techgr.c, but it reads all the vehicles and determines
a regional technology group distribution. The jointg.c (Figure 4.11) program brings
the separate zonal and ZIP code distributions using a control file (szzp.asc) containing
zonal ids and corresponding ZIP codes. It is important to adjust the code to the
address match success rate developed in the preprocessing. Currently the rate is set to
81% zonal and 19% ZIP code, suggesting that 81% of the vehicles were successfully
geocoded (the match rate for the test dataset). Jointg.c writes three ASCII files, one
for engine starts, one for running exhaust, and one for the SCF approach. Sz-tg.aml
reads the outputs of the C programs and converts the files to dbase files.
Few errors are generated with this module; however, many errors occur in the
pre-processing data development stage. These and other errors are discussed in
Chapter 4. As much as 40% of vehicle representation is lost due to decoding errors.
Errors resulting from the address matching process further degrade the spatial quality;
however, the use of distributions of groups may lessen the impact.
4.3.4. Major Road Technology Groups Module
The purpose of the major road technology group module is to estimate on-road
fleet distributions. The major road technology groups module reads the zone-based
retg.dbf, regiontg.dbf and scftg.dbf (Figure 4.12 shows the module). By utilizing the
sz and mr coverages, it develops road segment specific fleet distribution estimates.
This is accomplished in the mr-tg.aml process by determining a 'local' fleet
distribution and combining that with .a regional fleet distribution (Figure 4.13). The
local fleet is defined as an aggregation of vehicles registered within a 3 km radius of
the road segment. This process takes the longest of all of the modules because it has
to develop the combined distribution for each of multiple thousands of road segments
individually. The output of this process is a single dbase file called mrtg.dbf
containing the road segment id and the technology group percentages.
68
-------
Significant errors can occur in this process due to the unvalidated nature of the
assumptions. The assumption that the on-road fleet can be predicted by combining
distributions from a region and a 'user-defined' local fleet is loosely based on research
completed by Tomeh, 1996. Some evidence described in Chapter 6 is available that
defends the estimate.
4.3.5. Engine Start Activity Module
The engine start activity module predicts the number of engine starts by sz poly
and by hour of the day. Figures 4.14 and 4.15 represent the flow of the engine start
activity estimation process. This is accomplished by reading the TAZ trip generation
results file (taz.dat) and disaggregating the trips to sz zones using landuse, housing
units, zone size, and school / university landmarks. All home-based trip origins are
assigned to residential land uses based on household density. Shopping return trip
origins are allocated to commercial land uses based on area. University return trip
origins are allocated to zones that contain university landmarks based on enrollment.
Grade school return trip origins are allocated to zones that contain grade school
landmarks based on enrollment. Work return trip origins are allocated to non-
residential zones based on area. If conditions exist that do not allow allocation of
known trips using the above rules, trips are allocated to all zones based on area. Once
trips are allocated to sz polys, they are disaggregated to hours of the day using hourly
factors developed from Atlanta surveys for each trip purpose. The output of the
module is a dbase file of total trip origins by zone and time of day.
Error in this process is largely the result of poor input data quality. The use of
the travel demand forecasting values defines the highly aggregate original form of the
data. A major missing component is the inclusion of intra-zonal and external travel
patterns. Engine starts that result from individuals that go to destinations within the
TAZ will not be accounted for. Intra-zorial / external trips that originate outside the
study area and stop inside the site, will also not be represented.
4.3.6. Major Road Running Exhaust Activity Module
The running exhaust activity module determines hourly traffic conditions on all
major road segments found in the mr line coverage. Figures 4.16 and 4.17 illustrate
the process. Predicted conditions from the travel demand forecasting model output
file (tdfn.dat) are combined with an hourly factor to predict road segment specific
hourly volume, average speed, and LOS. The AML process also reads the grade
ASCII files, joins them into a point coverage, assigns each point to the closest major
road segment, and summarizes the grade points into five intervals. The data
summaries of grade, static conditions, and dynamic conditions are written to a dbase
file called mr-act.dbf storing results by the arid key field.
69
-------
Besides input data error, the incorporation of road grade causes some
significant problems. While the anticipated technique for collecting road grade (GPS)
results in absolute positional accuracy that exceeds the road database, it is the
relational accuracy that allows useful locational data to be collected. The process of
snapping points to the lines in order to develop the relational structure can produce
poor results around intersections and close parallel roads. Further analysis on this
subject is available in Chapter 6.
4.3.7- Minor Road Activity Module
The r ; nor roads activity module develops estimates of the mean travel time
and hourly trios occurring within each of the minor zones. Figures 4.18 and 4.19
illustrate the process. The shortest path is determined between the sz polygon
centroids and the closest node of the major road network. The aggregate travel time is
developed along with summaries of the hourly trip production at the zone (from the sz-
act.dbffile.). The final output is a dbase file called mz-act.dbf containing an aggregate
travel time value and hourly trip production for each minor zone.
The potential errors in estimating the travel time are substantial. However, the
alternative use of centroid connectors or zonal surface area would not allow a measure
of network configuration within the zone.
4.3.8. Engine Start Emissions Module
The engine start emissions module predicts hourly CO, HC and NOx emissions
or engine starts in sz polygons. Figures 4.20 and 4.21 illustrate the process flow.
generally, engine start technology groups are combined with emission rates and
estimates of the number of hourly trips to develop the emission estimates. Output is
written to a dbase file and stored in the em directory.
Emissions are elevated at the start because catalyst control equipment needs to
operate at high temperatures. While the actual emissions are dispersed as the vehicle
travels for the first few minutes, the model allocates the entire start portion to the
origin sz. Therefore, spikes of high emissions may be estimated at high population
density locations when in actuality the emission production at that location would be
lower. This source of error in position is significant. Although the 'puff allocation is
significant in identifying the original sources of high start emissions, the actual
location that the emissions entered the atmosphere is misrepresented.
70
-------
4.3.9. Minor Road Running Exhaust Emissions
The minor road running exhaust module has the task of predicting hourly
emissions of CO, HC, and NOx, given the activity conditions provided by mz-act.dbf.
Figures 4.22 and 4.23 show the module flow. While information regarding technology
groups is available for minor zones, traffic flow and modal conditions are not. Speed
and acceleration data have not been collected to provide clues for determining local
road profiles. For this reason, the module uses an average high and normal emitter
running exhaust emission rate for the three pollutants of concern. As soon as data
become available, the minor zone emissions can follow a similar track as the major
road emissions model.
4.3.10. Major Road Running Exhaust Emissions
The major road running exhaust emission's module calculates an aggregate
hot-stabilized and enrichment emission estimate for all major roads. The module is
represented in Figures 4.24 and 4.25. Estimates of vehicle activity and on-road
technology groups are combined with technology and operating mode specific
emission rates to develop hourly, road segment estimates of CO, HC, and NOx.
Matrices of modal variables are available as lookup files. These files match the format
of the speed / acceleration profiles, and consist of zeros and ones. The two files are
multiplied together, and the resulting speed and acceleration file bins are summed.
The values represent the fraction of activity in that particular mode.
4.3.11. Gridded Emissions
The gridded emissions module develops the total emission estimates by grid
cell in a raster format. It has the task of overlaying a user-defined grid polygon
coverage with the sz, mr, and mz coverages, and aggregating all the emission estimates
(weighted by the area or length). All of the emission mode estimates (engine start,
minor zone, major road, and SCF) are converted to raster datasets. The engine start,
minor zone running exhaust, and major road running exhaust are summed together to
develop estimates of total hourly CO, HC, and NOx. The process for obtaining gridded
emissions is shown in Figures 4.26 and 4.27.
4.4. Conclusion
Chapter 4 described how the applied model design from Chapter 3 was
translated into a functional computer model. Detailed descriptions of the input files
and system structure are provided. Flow charts for the various modules and programs
71
-------
are provided as well. The model includes both SCF emission rates and modal
emission rates. Engine starts and running exhaust are calculated separately for normal
and high emitting vehicles. Gridded, hourly emission estimates are produced.
72
-------
census
I - landuse
lOCATWN(p<>ty}~
WCATION(poly)
TZ1D
sz
LOCATION(poly)
CBID - -
izm •
iSZID
ZIP
ZIPcode
tOCATlQN(poty)
ZIP
sz.dat
CBID
LU
TZID
SZID
socioeconomic
data
Figure 4.2 - Zonal Environment Entities
73
-------
Start-
Census Blocks
Polygons ',-
' land Use
'Polygons
TAZ Polygons
Spatial Join
(100 meter fuzzy
tolerance)
/Start Zone. /
Polygons /
Create SZ Data File and
Calculate Housing Units
and Land Use Areas
SZ INFO
Data File
End
Figure 4.3 - Zonalenv.aml Flow Chart
74
-------
ITJD
ifcRID
Figure 4.4 - Road Environment Entities
75
-------
Minor Road Running
Exhaust Emission,/
Zone Polygons ,
End
Figure 4.5 - Roadenv.aml Flow Chart
76
-------
address
geocoding
zone.hne
CBID
VIN
vehicle characteristics
zip.tine
ZIPCODE
VIN
vehicle characteristics
hne2iw[.c
zone.tw t
CBID
VIN
vehicle characteristics
zip iu i
ZIPCODE
VIN
vehicle c hann lenslit s
Figure 4.6 - Vehicle Characteristic Entities
77
-------
zone.twt
CBID
VIN
vehicle characteristics
ZIP.twt
ZIPCODE
VIN
vehicle characteristics
zones. tg
CBID
frequency
con%
cohft
hcn%
h'ch%
non%
noh%
zonre.tg
CBID
frequency
con%
coh%
hcn%
hchft
non%
noh%
ZIPes.tg
ZIPCODE
frequency
con%
coh%
hcn%
hch%
nonft
noh%
ZIPre.tg
ZIPCODE
frequency
con%
coh%
hcn%
hch%
non9c
noh%
zonscf.tg
CBID
frequency
my%
ZIPscf.tg
ZIPCODE
frequency
my%
es.ig
CBID
frequency
hchtf
re.lg
CBID
frequency
hcn9t
hch%
non9t
scf.tg.
CBID
frequency
my*
Figure 4. 7 - Technology Group Entities
78
-------
Registration ascii file
with VIN and
address information
1
Address match
registration data
, j
/
Running exhaust
em issions road lines
wilb addresses
yes
Successfully
matched records,
wilh VINs
Unsuccesfully
matched records
Figure 4. 8 - Address Matching Flow Chart
79
-------
Temporary
ZIPcode Vehicle
File
Temporary Start
Zone Vehicle File
Figure 4. 9 - Vehicles.mak Flow Chart
80
-------
Start Zone
Vehicle Fife
Start Zone
Technology
Group (TG) File
ZIPcode
Vehicle File
ZIPcode
Technology
Group (TG) File
Start Zone Start
TG Distributions
Start Zone
Running Exhaust
TG Distributions
ZIPcode Running
Exhaust TG
Distributions
ZIPcode Start
TG Distributions
End
Figure 4. 10 - Techgr.c Flow Chart
81
-------
ZIPcode and start zone
ID file an match rate
(ie.,SZ=0.8,ZIP=0.2)
Start Zone Engine
Start TO
Distributions
Start Zone
Running Exhaust
TC Distributions
ZIPcode Engine
Start TO
Distributions
Determine Engine Start
and Running Exhaust
TG Distribution Based
on Match Rate
Start Zone Engine
Stan TG
Distributions
ZIPcode Running
Exhaust TG
Distributions
Stan Zone Running
Exhaust TG
Distributions
SCF Running
Exhausl TG
Distributions
Figure 4.11 - Jointg.c Flow Chart
82
-------
LOCATlQN(line)
TFID
ARID
LOCATlON(poly)
CB1D
TZID
SZID
lOCATlQWpoly)
CBID
frequency
con%
coh%
hcn%
hch%
non7c
notice
Figure 4.12 - On-Road Technology Group Entities
83
-------
Running exhaust
TG file by start
zone
Regional Running
exhaust TG file
Find all households
within 3000
meters
/Start zone
polygons
IRoad segment
specific local running
TG distributions
Figure 4.13 - Mr-tg.aml Flow Chart
84
-------
taz
iOCATION(poiy)
tZID ; '-%V;"v
trips by type
sz.dat
SZID
TZID
landuse
household data
landmarks
schools
Universities
temporal.lu
hourly
distributions by
trip type
sz-act.dbf
SZID
hourly trips
Figure 4.14 - Start Zone Activity Entities
85
-------
Start
Determine land use
fractions and total
housing units for each
TAZ
Disaggregate TAZ HB
trip productions to
residential SZs weighted
by housing unit density
Disaggregate TAZ HBW,
HBOTH trip attractions
and all NHB trips to
non-residential SZs
Disaggregate TAZ HBSH
trip attractions to
commercial SZs
Disaggregate TAZ HBGS
trip attractions to SZs
w ith grade school
landmarks
Disaggregate TAZ HBU
trip attractions to SZs
w ilh university
landmarks
Determine hourly
trip origins by SZ
7
7
SZ data file
7
SZ activity
file
Dbase file of szid,
and hourly trips
End
Figure 4.15 - Sz-act.aml Flow Chart
86
-------
grade. xy
G-ID
X
y
•^•^ ,• -;- -^-^s^^^-^^aji^^^
l-:,,,*^nTO^jf -:vg|/.
lOCATlONdine
JF-ID V«>
RD-JDC "
jSDrlDv'.-^V*
K • * . - -i ^
feA, ^J *xit±- .^,.^^,t,.»jSJ ^ JJR4J
\
\
>.
grade.gr
G-ID
grade
time
flag
^ >
mr-ac
>
f ^
t.am 1
f
mr-act.dbf
ARID
geometries
hourly volume
hourly v/c
hourly avg speed
tdfn.dat
TFID
predicted link
level traffic
parameters
/
/ temporal.factors
/
hourly on-road
factors
^
^
Figure 4.16 - Major Road Activity Entities
87
-------
Start
Grade location
file (ID,X,Y)
7
Grade attribute file
(grade, time,
quality)
Major roads
Travel demand
forecasting
network data
/ Hourly on-road
activity data
Generate grade
observation points
Grade pomts
with attributes
'Snap' grade points to
road segments with
100 meter tolerance
Develop static
geometric
characteristics for each
road segment
Develop hourly
dynamic traffic flow
parameters of volume,
v/c, and LOS
Grade points.
Major road
vehicle activity
data
Major road
activity dbase file
End
Figure 4.17 - Mr-act.aml Flow Chart
88
-------
. allroads
LOCATWN(line)
ARID
classifications
sz-act.dbft
SZID
hourly trips
mz-act.aml
mz-act.dbf
MZID
mean travel time
hourly trips
sz
LOCATlON(poly)
C.BID.,
LUID
TZID
SZID '
Figure 4.18 - Minor Road Activity Entities
89
-------
All road lines
Find euclidean distance
from centroid points to
closest non-interstate
road node
Determine average
travel time to major
road using 25 mph
speed
dbase file of mzid,
hourly Irips, and
mean travel time
Figure 4.19 - Mz-act.aml Flow Chart
90
-------
sz-em .dbf
SZID
hourly CO
hourly HC
hourl\ NOx
Figure 4.20 - Start Zone Emissions Entities
91
-------
Start
Start zoneTG
distributions
Calculate total start
emissions (total trip orgins
x TG distributions x TG
specific emission rate)
Start zone hourly
trip origins
Start emission rates by
TG for CO,HC,NOxin
grams per start
Hourly start zone engine
start emissions of
CO.HC, and NOx in
grams
End'
Figure 4.21 - Es_emission.c Flow Chart
92
-------
allroads
10 CAT! ON (line)
ARID
classifications
sz-act.dbft
SZID
hourly trips
mz-act.aml
mz-act.dbf
MZID
mean travel time
hourly trips
sz
lOCATWN(poty)
CBID
LUID
IZID
SZID
Figure 4.22 - Minor Zone Activity Entities
93
-------
Starts
Minor road running
emission zone-ids and
total hourly travel time
Calculate hourly
emissions using
constant normal and
high emitter rates
Hourly minor road
emissions of CO, HC,
and NOx in grams
End
Figure 4.23 - Mz-em.aml Flow Chart
-------
Figure 4.24 - Major Road Emissions Entities
95
-------
Hourly, major road
running exhaust
emissions of CO, HC
and NOx in grams
Figure 4.25 - Re_emissions.c Flow Chart
96
-------
mz-em.dbf
MZ1D
hourly CO
hourly HC
houri\ NO*
raster data
lotal emissions
engine start emissions
minor road running exhaust
major road running exhaust
SCF emissions
Figure 4.26 - Gridded Emissions Entities
97
-------
SZ potyg on s
7
MZ polygons
Hourly, start zone
engine start emissions
of CO.HC, and NOx in
grams
Hourly, major road
running exhaust
emissions of CO, HC,
and NOx in grams
Sum all emissions by
grid cell and weight
by area or length
Hourly gridded
emissions of CO,
HC, and NOx
Hourly minor road
emissions of CO, HC,
and NOx in grams
Raster files of engine starts,
minor road running exhaust,
major road running exhaust,
and SCF running exhaust
emissions
End
Figure 4.27 - Grid-em.aml Flow Chart
98
-------
5. MODEL DEMONSTRATION
This chapter demonstrates model capabilities. By applying the model
described in Chapter 4 for a study area, an estimation of emissions will be developed.
While the conceptual value of the model is revealed through its design parameters, a
demonstration provides insight into the model's practical value. The model will
predict grams of CO, HC, and NOx, with a spatial resolution determined by the user.
In this case, hourly gridded estimates are provided for 100 meter, 250 meter, 500
meter, and 1km grid cells.
The study area is a 100 square kilometer portion of the Atlanta, Georgia
metropolitan area and is shown in figure 5.1. The 10 kilometer by 10 kilometer slice
of the northeast suburbs was selected as a sample study area because it contains:
diverse landuse, variable densities of development, a major interchange, major north-
south arterials (leading to the CBD) and an interstate known for congestion (northern
portion of 1-285).
The following input datasets were used:
• 7995 Georgia Department of Motor Vehicles Registration Dataset
• 1990 US Census Summary Tape File 3a
• 7994 US Census TIGER File
• 1995 Updated TIGER Road Database
• 1995 Atlanta Regional Commission's (ARC) Traffic Analysis Zones
• 1995 ARC's Travel Demand Forecasting Network
• 1995 ARC's Land Use Data
• 1996 ARC"sARCMAP Road Database
99
-------
Study Area
10 km by 10 km
10 0 10 20 Kilometers
A
Figure 5.1 - Model Study Area Site Map
100
-------
5.1. Preprocessing
Several preprocessing steps were completed to prepare model input data. Most
model implementation efforts will require substantial preprocessing steps due to the
variability of data availability. In this sample case, preprocessing was needed for the
road network and vehicle characteristics development. Other data required simple
conversion or transformation. Several preprocessing steps were needed.
5.1.1. Vehicle Characteristics
The Georgia DMV Registration Database is protected under privacy
regulations. The Georgia Tech Air Quality Laboratory (AQL) has data access
permission for research (under contract). To protect the privacy of vehicle owners, a
three step 'double-blind' procedure was used to provide vehicle data. First, data
consisting of owner address information and a unique identifier were transferred from
the AQL. Second, the data were address-matched (in the GIS), and aggregated to
Census Block or ZIP code. A file of the unique identifier and the zonal identifier was
transferred back, to AQL. Third, a file of the zonal identifier and vehicle identification
number (VTN) was returned from AQL, thereby providing a spatially-resolved,
decodable, file of vehicles.
Address-matching provides the ability to develop vehicle registration
information at a better spatial resolution than provided by ZIP codes. The vehicle file
was address-matched using two road datafiles, the ARC's ARCMAP Road Database
and the road database. The ARCMAP database provided comprehensive coverage for
the entire metropolitan Atlanta area. The road database provided higher spatial
accuracy, but did not cover the entire area. The road database was used first, to
maximize spatial accuracy, and then the ARCMAP was used to maximize
comprehensiveness. Actual vehicle locations were offset by 30 meters to ensure that
vehicles would not fall on zonal boundaries when aggregated. For a successful match,
the ZIP codes must match. Slight errors in spelling were allowed. The road database
resulted in a 63% match rate. Another 18% were matched using the ARCMAP road
database, for a total of 81%. Therefore, two files were created; one of matched
vehicles (81%) and one of unmatched vehicles (19%) (discussed further in chapter 6).
The matched vehicles were aggregated to US Census Block (census) polygons as
described in the previous paragraph.
The two files were sent to the vehicles.mak PC process [see figure 4.6]. This
process was developed by John Leonard, William Bachman, and Osama Tomeh- at
101
-------
Georgia Tech in 1996 [Tomeh, 1996]. The process reads a file consisting of a record
identifier and VIN. During the process, the VIN is decoded using software developed
by Radian International Corporation,jthe vehicles emission test weight was added
using a lookup table, vehicles are flagged'as being high or normal emitters, and
emission-specific characteristics are written to an output file (record identifier, vehicle
characteristics, emitter types). Each of the files was sent through this process resulting
in the two files required as inputs to the emissions model.
5.1.2. Conflation
Conflation is the process of combining two separate line datasets into a single
dataset. In the model, the prognostic data (volumes and speeds) provided by the travel
demand forecasting network are transferred to a road network that has better spatial
accuracy. There is not a one-to-one correspondence between the two datasets' road
segment representation, nor are there attribute fields that can create connectivity. The
abstract spatial structure of the travel demand forecasting network prevents a clearly-
defined locational connectivity. However, enough spatial definition exists that a
manual link-by-link assessment can establish connectivity. The process of conflation
is frequently used by transportation agencies to bring various linear datasets together.
The sample area's portion of the travel demand forecasting network consisted
of 532 links, and the accurate road database consisted of 3602 road segments.
Overlaying the datasets in the GIS (ARC/INFO) identified representational
similarities. As individual travel model links were 'selected', corresponding accurate
road segments were also selected. The NAVTECH roads were assigned an identifier
field that could be used to transfer attributes (predicted volume, speed, etc.). Each of
the 532 was processed in this manner. The resulting database is one of the required
inputs into the model.
5.1.3. Other Steps
Numerous other steps were needed to fully prepare data for model running.
The databases are all distributed in different formats, requiring conversion,
transformation, and renaming. Atlanta area Census data (STF3A and TIGER) had to
be selected, joined, and transformed to develop the zonal database containing detailed
household information. The ARC's travel demand forecasting network had to be
converted from an ASCII file to an ARC/INFO coverage using programs written by
Wayne Sarasua, Xudong Jia, and William Bachman at Georgia Tech. The ARC's
TAZs and land use were delivered in an ARC/INFO coverage. Other urban areas may
have to develop customized strategies to get the input information in the format
described in the previous chapter, and in the data dictionary found in the appendix.
102
-------
5.2. Spatial Environment
The spatial environment consists of the ARC/INFO input coverages: taz
(Atlanta Regional Commission's (ARC) traffic analysis zones), census (US Census
blocks), landuse (ARC's landuse), ZIPcode, and allroads (conflated road database).
The spatial environment output coverages were: sz (engine start polygons), mr (major
road running exhaust lines), and mz (minor road running exhaust polygons). Figures
5.2 and 5.3 demonstrate the'connectivity.
SZ
(SZID
CBID
TZID
LU)
Figure 5.2 - Engine Start Zone Creation
103
-------
Travel Demand
Forecasting Model
(TFID)
CONFLATION
k
Accurate
Road Database
(ARID)
- lines
(MRID
TFID
ARID)
MZ polys
(MZID)
Figure 5.3- Running Exhaust Entity Creation
5.2.1. Engine Start Polygons
Engine start polygons (SZ) consist of spatially joined features from taz, census,
landuse, and zipcode. The 1624 SZ polygons had a mean area of 72,746 square
meters. Most of the polygons are identical to smallest input polygons, generally, the
US Census blocks. Each SZ polygon maintains identifiers to the original databases in
a one-to-one or many-to-one relationship
104
-------
5.2.2. Running Exhaust Lines and Polygons
Running exhaust lines and polygons (MR and MZ) consist of conflated travel
demand model network segments and areas bounded by those segments. The
segments represent roads that have prognostic estimates of travel behavior from the
ARC. The polygons represent the roads that have road-specific estimates of travel
activity. The road segments had a median length of 202 meters. The median of the
205 minor zones is 487,815 square meters.
5.3. Fleet Characteristics
The fleet characteristics were developed from the two files described in section
5.1.1. Five fleet distribution databases were created by the model from the two files;
es.tg (zone-based engine start technology groups), re.tg (zone-based running exhaust
technology groups), scf.tg (zone-based SCF technology groups), rereg.tg (regional
running exhaust technology groups) and mr.tg (major road running exhaust technology
groups). The files contain identifiers that connect records to spatial entities (sz or mr).
5.3.1. Model Year Distributions
The distributions of vehicle model years predicted by the model for the sample
area are shown in Figure 5.4. The figure shows the entire sample area's mean
distribution and two sample sz polygon distributions. The sample size for the zone
with SZID of 2176 was 109 vehicles and SZID 209 was 126 (zonal frequencies varied
from zero to several hundred). The entire area's mean frequency was 49. In Chapter
6, a comparison between observed and estimated model year distributions is presented.
0.16
SZID=2176
SZID=209
Area Mean
Figure 5.4 - Model Year Distribution
105
-------
5.3.2. High Emitters
High emitting vehicles were defined in one of the preprocesses [see section
3.4.2.3]. -In the process, vehicles are randomly selected from four groups of vehicles,
each having different likelihoods of being high emitters. The 'flagged' vehicles are
then characterized as high emitters. The resulting sample area high emitting vehicle
distributions are shown in figure 5.5. The figure shows a dot density map of all engine
starts, CO high emitters, HC high emitters, and NOx high emitters. The two circles
identify locations with high numbers of engine starts, but low numbers of high emitter
starts, suggesting that the likelihood that a vehicle is a high emitter, varies spatially.
Although this can't be validated until other model components are validated, it does
suggest a possible value of the model.
5.4. Vehicle Activity
The sample area vehicle activity was developed from the Atlanta Regional
Commission's (ARC) Travel Demand Forecasting Model dataset. Supplemental
information came from 11 speed and acceleration profiles and temporal factors from
Parsons Brinkeroff Inc.
5.4.1. Engine Start Activity
Engine starts were developed from trip generation data-at the ARC's traffic
analysis zone level. Trips were disaggregated to SZ coverage polygons based on the
ARC's 1995 land use data, 1990 US Census STF3A data housing unit densities, 1994
US Census TIGER data, and school and university landmarks developed by Georgia
Tech. The AM peak hour engine starts spatial distributions is shown in figure 5.5.
The engine start temporal distributions were developed using half hourly
distributions by trip type (home-based-work, home-based-shopping, home-based-
other, home-based grade school, home-based-university, and non-home-based).
Figures 5.6 through 5.10 show the hourly distributions of the trip origins, directly
translated to em-:ne starts. The non-home-based are not disaggregated by origin and
destination bt use there is no information regarding the origins or destinations.
5.4.2. Run iing Exhaust Activity
Running exhaust activity was estin >i for major roads and minor roads.
Minor roads consisted of all roads not expliciuy modeled in the ARC's travel demand
forecasting model. Major road were explicitly modeled by ARC for daily activity
using The Urban Analysis Group's TRANPLAN product. The network used in the
TRANPLAN model was conflated to accurate roads.
106
-------
Minor road running exhaust activity was predicted using the engine start
activity estimates for each SZ zone, and the shortest network path from the centroid of
that zone to the closest MR line. An average travel time for each path was determined
using an average travel speed of 30 MPH. The aggregate travel .time for activity in
each MZ became the estimate of minor road vehicle activity.
Major road running exhaust activity was developed directly from the
TRANPLAN output, and associated speed and acceleration profiles. Figure 5.11
represents the spatial distribution of peak hour volume density. Figure 5.12 shows the
temporal distribution used to divide daily activity into hourly segments. Eleven speed
and acceleration profiles were used to predict modal activity; five for each LOS on
interstates, five for each LOS on interstate ramps, and one for all other roads. The data
for one speed and acceleration profile used for the non-interstate, non-ramp, roads
were collected on a major arterial between signalized intersections.
Road grade data were available for approximately 25% of the interstates
running through the sample area. All roads without grade information were assumed
to have a zero grade. Roads with grade information had the grade distribution
segmented into five intervals.
5.5. Facility and Gridded Emissions
The emissions estimates for the sample area were developed for engine starts
and running exhaust activity. Engine start emissions were developed for each sz
polygon. Running exhaust emissions were developed for mz polygons and mr lines.
All estimates are in grams. The gridded estimates were aggregated from the facility
entities at 100, 250, 500, 1000 and 2000 meter grid cell sizes (the value refers to the
length of one cell side, not the cell's area). The various sizes were developed to
explore and demonstrate the impact of grid cell size on the emissions estimate.
Figures 5.13 through 5.19 represent the 100 meter grid cell aggregation in a surface
(with an overlay of 500 meter grid lines). Figures 5.20 5.22 show temporal
distributions of emissions.
5.5.1. Engine Start Emissions
Engine start emissions are shown in figure 5.13. The figure shows a three-
dimensional surface where x and y are geographic coordinates and z is adjusted value
for engine start emission estimate. The value is adjusted to spatially identify relative
emission estimates. The individual 'spikes' show estimated high concentrations of
engine start emissions. Given the hour of the day (7-8 AM) the majority of the
•emissions from engine starts occur in residential areas. Thus, areas with large
107
-------
populations (resulting in large numbers of engine starts) combined with higher
emitting vehicles spatial distributions will have high emissions. In the figure, the
highest spikes in the north east corner of the map are locations of dense multi-family
development.
The total 24-hour engine start CO estimated for the area is 15,768,000 grams,
the total estimated HC is 347,000 grams, and the total estimated NOx is 571,000
grams.
5.5.2. Minor Road Running Exhaust Emissions
Minor road running exhaust emissions are shown in figure 5.14. The same
format as described in section 5.5.1 is used for the figure. As seen from this figure,
spikes of intense emissions are not prevalent. This is due to the highly aggregate
nature of estimating local road activity to large polygons. However, variability is
evident among zones, indicating the impact of the road network configuration on
travel time.
The total 24-hour minor road running exhaust CO estimated for the area is
1,195,000 grams, the total estimated HC is 50,000 grams, and the total estimated NOx
is 60,000 grams.
5.5.3. Major Road Running Exhaust Emissions
Major road running exhaust emissions are shown in figure 5.15. As expected
emission 'spikes' fall along the major road (shown as white lines). The highest spikes
fall along the interstates. Arterials, especially the Peachtree Industrial Blvd Peachtree
Road arterials, show significant emissions as well. The emissions estimates are not
linear with volume, as appears, but are affected by the predicted modal behavior.
When speed and acceleration profiles are used that have high variability (arterials, or
low LOSs), there are higher emissions.
The 24-hour major road total running exhaust CO estimated for the area is
18,375,000 grams, the total estimated HC is 734,000 grams, and the total estimated
NOx is 811,000 grams.
5.5.4. SCF Running Exhaust Emissions
Figure 5.16 represents speed correction factor (SCF) running exhaust
emissions. Again, the emission spikes fall along the major roads. However, the
emissions along arterials do not appear as significant as the previous figure. Mostly,
this is due to the use of average speed as the measure of vehicle activity, excluding the
effects of variable acceleration and deceleration. Thus, arterials and poor LOS road
segments may have poorly represented emission estimates. While the level of spatial
108
-------
aggregation used for reporting (4-5 km grid cells) may negate this impact, it is clear
that facility-level and smaller grid cell aggregations of emission estimates will be
affected.
The total 24-hour SCF running exhaust CO estimated for the area is
17,045,000 grams, the total estimated HC is 1,236,000 grams, and the total estimated
NOx is 4,432,000 grams.
5.5.5. Total Emissions
The total emissions estimates were developed by adding the 100 meter
aggregations of engine starts, minor road running exhaust, and major road running
exhaust (aggregate modal). Total emissions for the study area are represented in
figures 5.17 to 5.22. Figures 5.17 to 5.19 show emissions estimates for CO, HC, and
NOx, between 7 and 8 AM. Figures 5.20 to 5.22 show the temporal variability found
in the estimates occurring between 6 AM and 9 PM. The CO and HC estimates are
characterized by the major road emissions and the spikes of engine start emissions.
The NOx estimates are characterized by emissions on the major roads.
The figures showing temporal variability identify the impacts of the
distributions seen in figures 5.6 through 5.12. It appears from the figures, that engine
start emissions dominate the off-peak emissions, while running exhaust emissions
dominate the peak hour emissions. This may be a function of the impact of congestion
on the roads (higher variability in modal activity) during high traffic times. There also
appears to be an unusual amount of engine start activity between 3 and 4 PM.
Reviewing the temporal curves, all trip types except home-based-work are high
between 3 and 4 PM.
5.6. Conclusion
Chapter 5 presented the results of model runs on a 100 sq. km area in
northeast Atlanta, GA. The results of each module are presented along with detailed
descriptions of the input data. Outputs are described in detailed as well.
109
-------
n
CO HC NOx
High Emitter Engine Starts 1-8 AM
(1 dot = 1 start)
All Engine Starts, 1-8 AM
( 1 dot = 20 starts)
Figure 5.5 - High Emitter Engine Starts, 1-S AM
u.*t -
0.35 -
0.3 -
0.25 -
0.2
0.15 •
0.1 -
0.05 -
0 -
, , _ n
E E E
-i
•-;
pi
1
1
n H •
JJn_r^-*TcH-
.
DHBW(towork)
• HBW (to home)
EEEEEEEEE
CO CO CO
T ?? "?
CNJ c\i -4
Figure 5.6 - Home-Based Work Trip Temporal Distribution
110
-------
0.14
0.12 -•
0.1 -
0.08
0.06
0.04
0.02
0
EEEEE
CO CO CO CO CD
D HBSH (to shop)
• HBSH (to home)
1
E
Q.
E
Q.
CO
O 1-
Figure 5.7- Home-Based Shopping Trip Temporal Distribution
0.6 -
0.5 -
0.4 -
0.3 -
0.2 -
0.1
n
E E E
CD CD CO
-,
CO
-]
1 1 t-wn __ __|
EEEEEEEE
CO CO Q- Q- CL CL Q. O.
•^coLnt^a>-^--^cninr^a>i-
D HBGS (to school)
• HBGS (to home)
Figure 5.8 - Home-Based Grade School Trip Temporal Distribution
111
-------
Figure 5.9 • Home-Based University Trip Temporal Distribution
NHB
0.1 -
0.08
0.06 •
0.04 -
0.02 -
0 -
E E E E
CO CO CO CO
•^ co un N
CNJ CM ^r CD
(i
—
_f
?
p
-
—
.n,r-i,n
EEEEEEEE
CO CO Q. Q. Q. Q. Q. Q.
O>T-T- COLO r-^ o> T-
COVCMCAjTtCDCo'T
D NHB
Figure 5.10 - Non-Home-Based Trip Temporal Distribution
112
-------
1-285
Peachtree
Road
Peachtree
Industrial Blvd.
Interchange
1-85
Figure 5.11 - Road Volume Density
0.18 -
0.16 -
0.14 -
0.12 -
0.1 -
0.08 -
0.06 -
0.04 -
0.02 -
0 -
_
_n
i—i
llnnnnn
"
pi
-
nnnfl
Don-road
E E E
CO 03 CO
••- co in
C\J CM ^T
E
OJ
E E E
03 CO Q.
O) ••- t-
CD
CM
O i-
E
Q.
op
C\l
E
Q.
Q.
h-
CD
a
cp
ab
Figure 5.12 - On-Road Activity Temporal Distribution
13
-------
1-285
Peachtree
Industrial Blvd.
1-85
Figure 5.13 - Engine Start CO, 7-8 AM
Grid Cell = 0.5 km
114
-------
1-285
Peachtree
Industrial Blvd
1-85
I - 285 /1 85
Interchange
10 km y 10 km
Grid Cell = 0.5 km
Figure 5.14 - Minor Road Running Exhaust CO, 7-8 AM
115
-------
Peachtree
Industrial Blvd
1-85
1-285
N,
Figure 5.15 - Major Road Running Exhaust CO, 7-8 AM
I 285 /1 85
Interchange
10lfm Y in Vrn
Grid Cell = 0.5 km
116
-------
1-285
Peachtree ••
Industrial Blvd.
1-85
I - 285 /1 85
Interchange
v 10 km
Grid Cell = 0.5 km
Figure 5.16 - SCF Running Exhaust CO, 7-8 AM
117
-------
Peachtree
Industrial Blvd.
1-85
1-285
Figure 5.17 Total CO, 7-8 AM
Grid Cell = 0.5 km
118
-------
I 285
Peachtree
Industrial Blvd
1-85
N,
I 285/185
Interchange
10 km Y in Ion
Grid Cell = 0.5 km
Figure 5.18 - Total HC, 7-8 AM
119
-------
1-285
Peachtree
Industrial Blvd.
1-85
I - 285 /1 85
Interchange
10 km x 10km
Grid Cell = 0.5 km
Figure 5.19 - Total NOX, 7-8 AM
120
-------
6-9 AM
10 AM - 1 PM
- ^ X
2 - 5 PM
'.i i'
6 - 9 PM
/I-
Figure 5.20 - Total CO, 6 AM - 9 PM
-------
6-9 AM
/T
10 AM -1 PM
2-5PM
6 - 9 PM
/.
Figure.5.21 - Total HC, 6 AM to 9 PM
122
-------
6-9 AM
10 AM -1 PM
•.'- f
* ; •, ."j, /
2 - 5 PM
6-9PM
Vfc :/ •-
/ X &?.
/ . .! • -S(:-.,<
Figure 5.22 - Total NOX, 6 AM to 9 PM
123
-------
6. MODEL EVALUATION
This chapter evaluates the model by analyzing and discussing potential sources
of error, paying particular attention to the spatial data issues discussed in section 2.4.3;
positional accuracy, resolution, and content. Studying the sources of model spatial
error provides insight into developing validation studies and future research needs.
One of the model objectives identified in chapter 3 was for the model to be statistically
sound. Exploring the sources of error, and their propagation in the model, will also
help determine appropriate strategies for developing confidence bounds around the
spatially resolved estimates, an important model design feature. A sensitivity analysis
and a comparison between the aggregate modal approach and speed-correction-factor
approach are also provided.
A large amount of error in the model will be associated with the quality of the
input data. While input data error is not explicitly discussed, it should be evident that
any limitations associated -with the input data impact the model results. It should also
be noted that the input data's measures of spatial quality should focus on the relative
positional accuracies among the datasets, not just the absolute accuracy.
6.1. Spatial Environment
The spatial environment modules create the spatial entities si, mr, and mz.
Each entity was created by spatially manipulating input polygon and line data. During
the spatial manipulation, potential positional errors arise that would impact the
locational accuracy of the estimates. The following two sections describe the potential
issues.
6.1.1. SZ
SZ was created using polygon-on-polygon overlay techniques on the four
ARC/INFO coverages census, taz, ZIPcode and landuse (see figure 4.2). The
technique merges two or more polygon networks into a single polygon network. The
datasets share many common boundaries. However, the data were developed from
different sources and resolutions resulting in different representations of common
boundaries. In the sample area, the US Census blocks and the ARC's TAZs were
generated from the original' TIGER data, and, therefore, match very well. The ARC's
124
-------
land use and ZIP code data were developed from different sources. The impact of this
problem is that there is potential for misrepresentation of the landuse I TAZI census
combinations.
Figure 6.1 demonstrates the polygon-on-polygon overlay problem. The figure
shows a portion of the sample area's polygon structure. The left side shows census
polygons in black lines and landuse polygons in gray lines. The right side shows
census and landuse polygons in gray, and the resulting SZ polygons in black. Point A
shows a shared boundary between landuse and census that is represented differently.
Point B shows polygon boundaries that may, or may not, represent the same features; it
is too difficult to tell conclusively.
The polygon overlay process includes a 'fuzzy tolerance' that allows the user to
define a threshold that is allowable for matching edges. In the study model example, a
tolerance of 30 meters was allowed. The resulting polygon structure (SZ) is shown on
the right side of the figure. Point C shows the same area as point A, but the potential
'sliver' polygon was removed because the lines were within the tolerance level. Point
D shows borders that may have represented the same feature, but as the distance
between the two edges deviated more than 30 meters, they were represented as
separate entities.
Census and Landuse
Census, Landuse, and SZ
Figure 6.1 - Polygon Overlay Errors
As a result of the polygon overlay process, there are potential errors that exist
in the spatial representation of the joined polygons. There are three potential impacts
on the resulting data; data are lost, data are spatially misrepresented, and data
combinations are incorrect. The biggest risk to emission output quality is the loss of
data. By adjusting the boundaries of the spatial entities, any polygons less than 30
meters across would be removed, hi the sample study area used in Chapter 5, 5 of 925
US Census blocks were lost during this process. Most of the blocks were road
medians between divided highways and held no bearing on the emission estimates.
125
-------
Howevel^tone was a Census block that contained two households. In this case, two
houseiKdds have little impact, but other datasets may have more, suggesting that the
modelsuser would have to select a smaller tolerance level. The trade off to lower
tolerance levels is increased spatial misrepresentation.
Spatial misrepresentation and incorrect data combinations are difficult to
identify because the true values and positions must be known. Prior to model
operation, detailed quality control steps in data development will prevent further error
propagation. The model assumes that errors in input data exist, and therefore, a 'fuzzy
tolerance' is used that can be adjusted to minimize data loss and maximize accurate
representation.
6.1.2. MR and MZ
MR is created by selecting roads that are modeled in the travel demand
forecasting model. Spatial errors associated with MR occur in the preprocessing steps,
not in the formal model. The process of conflation, described in section 5.1.2,
involves a great deal of user input, adding an aspect of human error.
The biggest concerns resulting from conflation errors are missing roads and
miscoded roads. Matching some travel model links with the actual roads they
represent can be difficult because the travel demand model network consists of
abstract representations of roads. This is compounded by a lack of agreement or
reporting about road classification in multiple datasets. Further, commercially
available, accurate road datasets (similar to NAVTECH used in the study area) use
significantly more detail than the travel models. One travel model link usually
represents many road segments. During the conflation process, it is easy to miss one
of the small segments, resulting in a 'gap' in the new network.
MZ polygons are created by defining the MR lines as polygon boundaries.
There are no problems with the spatial accuracy of the polygons, other than those
mentioned for the MR road network. However, there is one issue that is worth
mentioning. The polygons are supposed to represent aggregations of local roads. The
process of creating the MZ polygons does not actually consider the locations of these
roads, but are defined as any polygon bounded by major roads (or the outside
boundary). Therefore, medians from divided highways become defined as MZ
polygons. While this poses no impact on the emission estimates, it can impact the
amount of time required for model operation.
126
-------
6.2. Vehicle Characteristics
The spatial errors associated with vehicle characteristics are significant and
worth detailed study. There are several broad assumptions made in developing the
spatially-resolved fleet distribution estimates. First, it is assumed that a vehicle's
registered address is its 'home' location. This has not been proven or studied in
previous research. Second, it is assumed that all the registered vehicles have the same
probability of being operated at any given time. This not the case, but little evidence
exists that justifies adjusting the fleet distribution to more accurately characterize
operating vehicles. Third, it is assumed that any road segment's operating fleet
distribution is composed of two groups of vehicles, a 'local' fleet and a 'regional'
fleet. While some evidence exists to back up the idea [Tomeh, 1996], many questions
remain about the specific definitions of the two groups of vehicles.
The potential negative impact of the above assumptions is reduced by
predicting aggregate distributions rather than individual vehicles. The actual number
of vehicles predicted at each entity, or whether the right vehicle is predicted at each
entity, is less important than the predicted distribution. The only vehicle information
used by the model is the fraction of each technology group at the zonal or road
segment level, not the frequency. The model is more concerned with accurately
characterizing the fleet, than accurately identifying the fleet.
Measures of the model's ability to predict the fleet distribution must come
from future validation efforts. However, data do exist that indicate some biases found
in the decoding process. Figures 6.2 and 6.3 show the degradation of the quality of the
fleet estimate as vehicle characteristics are determined in the model. The basis for
comparison (model year) comes from the raw vehicle registration dataset. As vehicle
identification numbers (VINs) are decoded, model year information is predicted.
Comparing the predicted and original model year distributions for the various datasets
shows where bias has been introduced into'the system.
Figure 6.2 shows the drop in the frequency of each model year for three steps
in the decoding process; the VIN decoder, the removal of non-autos, and the
assignment of vehicle weight. The VIN decoder operation results in a 7.7 % loss of
data because the VIN couldn't be decoded. Most of those vehicles are older than
1980, when VINs were not standardized among manufacturers. Two odd 'humps'
occur (frequency is overpredicted) in model years 1973 and 1978. After removing
non-autos, only the 1973 hump remains. Further study revealed that the VEN decoder
software was incorrectly assigning pre-1972 BMWs and Volvos as 1973 vehicles.
Adding the test weight to the vehicles (removing those without matches)
resulted in a substantial data loss (39%). It also appears that the data loss is biased by
model year, with 1988 vehicles underrepresented and 1995 and newer' vehicle
127
-------
unrepresented. Figure 6.3 shows the resulting distributions. The final distribution
shows the impact of the data loss on the fleet distribution. . Pre-1972, post- 1994, and
1988 vehicles are under-represented. Mid-1980s and early- 1990s vehicles are over-
represented.
6.2.1. Zonal Fleet
There are two concerns regarding the spatial allocation of the vehicles to zones;
incorrect assignment, and non-residential trip distributions. Vehicles can be
incorrectly assigned to zones because of address-matching problems. Non-residential
trips use the fleet distribution of the current zone, disregarding the fact that the actual
fleet distribution consists of vehicles originating from other locations.
The address-matching process can result in a small percentage of vehicles that
were incorrectly assigned to zones because of errors in the address, an issue that has
been well-documented in the literature. This problem is minimized in the model by
having stringent matching guidelines; the vehicles' ZIP codes must match the
candidate addresses' ZIP codes, the road types must match perfectly, and there can
only be one error in spelling or incorrect address prefix (north, east, etc.). Further, the
road 'dataset being used for address-matching could be missing new subdivisions or
developments. If the registration dataset is newer than the last road dataset update,
there could be vehicles that fail to match. However, the 'failed' vehicles are not
discarded, but assigned a location based on their ZIP code.
The zonal fleet is developed from two sets of files, an address-matched file,
and a ZIP code file (address match failures). To bring these two groups of vehicles
together, the relationship between the SZ polygons and ZIPcode polygons must be
identified. Each SZ is apportioned part of the ZIP code vehicles based on a
comparison of the areas of the two polygons. A zone could have 78 vehicle address-
matched within its boundaries, and an additional 10.3 vehicles assigned to it from the
ZIP code. Since the concern of the model is the distribution, there are no problems
that arise from non-integer frequencies.
Zones that do not have any address-matched vehicles are assigned the fleet
distribution of the ZIP code. While some problems remain, zones that have new
subdivisions will be assigned a fleet distribution that partially represents the vehicles
registered at that location.
issue regarding the fleet distribution of non-residential trips is not handled
well ir model. Since vehicles are not tracked during estimates of activity, there is
no me, inism for tying the origin fleet distribution to the destination. Unless the
destination lies in a zone or ZIP code with a fleet distribution that is similar that of the
origins, an incorrect distribution will be assigned. Given the dynamics of land use
development, there is strong indication that strong bias will exist. For example, the
128
-------
fleet distribution of vehicles leaving a commercial land use zone is assigned the fleet
profile of registered vehicles in that zone and ZIP code, not the trip origin zones of the
operating vehicles.
-All vehicles
- VIN decoded (all
vehicles)
- VIN decoded (autos
only)
- After weight lookup file
(autos only)
Model Year
Figure 6.2 - Model Year Frequencies
Model Year
- Regional
-VIN decoded (all
vehicle types)
-VIN decoded-(autos
only)
-Test weight lookup
(autos only)
Figure 6.3 - Model Year Fraction
129
-------
6.2.2. On-road Fleet
Problems associated with the on-road fleet distribution estimation stem from
the unvalidated assumption that the on-road fleet can be summarized by combining a
local fleet (defined as all vehicles within 3 km) and a regional fleet (all vehicles in the
region). The instability of this approach is demonstrated by analyzing two sets of
observed on-road vehicle datasets. Figure 6.4 shows registered vehicle locations for
vehicles that passed through data collection sites in the study area. These data were
collected and provided by the School of Earth and Atmospheric Sciences at Georgia
Tech. License tags were captured on passing vehicles and matched to a registered
vehicle's VTN and address. At site A, 674 vehicles passed through the data collection
site. The figure indicates that spatial variation exists in the estimated origins of the
observed vehicles. However, the size and shape of the spatial variability are unclear.
Similarly, site B, with 13,481 vehicles, indicates spatial variability. The model
currently uses a 3 km radius to define a 'local' fleet (10% of the observed vehicles fell
in that range). It may be more accurate to select an alternative geometric (wedge, oval,
network distance, etc.) search pattern involving road types, time of day, and network
structure.
Site A: Arterial, 641 Vehicles
Site B : Ramp, 13481 Vehicles
Figure 6.4 - Observed On-Road Vehicle Origins
6.3. Vehicle Activity
Spatial errors associated with the estimates of vehicle activity can be tied to
previously mentioned problems with the spatial environment and the travel demand
130
-------
model limitations. The model shares the problems associated with the use of the travel
demand forecasting models in predicting emission-specific vehicle activity; inaccurate
speeds, no feedback into the distribution phase, etc. (see section 2.3.1). These known
travel demand model result limitations will not be discussed. However, trip
disaggregation, the use of regional temporal distributions, and speed and acceleration
matrices create some errors in vehicle activity estimates that are worth mentioning.
The disaggregation of trips by purpose to different landuse makes the broad
assumption that the landuse data are discrete. All home-based trips are assigned origin
engine starts to residential area. If homes can be found in other land uses, their engine
start activity estimates will not be assigned to the correct location. The landuse data
must be discrete to prevent this from occurring.
The use of regional temporal factors to distribute zonal and road segment
activity results in errors. A series of spatial queries using 1990 Census data in Atlanta
indicated that the fraction of people traveling to work between 6:30 .and 7:00 AM was
approximately 9% for people living within 2 kilometers of the central business district
(CBD). and about 15% for people living between 8 and 10 kilometers from the CBD.
By using regional temporal factors in the model, all zones and road segments are
assigned the average, not allowing spatial variability. Thus, the peak hour 30 miles
from the CBD will be the same as the peak hour 5 miles from the CBD.
The speed and acceleration profiles were used as a post-processor to the travel
demand model's output. They are used to predict the modal distributions of the
vehicles operating on the road. The model only includes matrices for interstates and
ramps, forcing all lower classifications to rely on a single profile of mid-block
estimates. As soon as new data are collected, validated, and available, the model
structure can incorporate the new findings. The impacts of modal activity around
signalized intersections could have a tremendous impact on the spatial variability of
the estimates. As is, running exhaust emission estimates are highly correlated to
volume. The potential variability found in future matrices could show that the highest
emissions occur around major intersections, not high volume, low modal variability
interstates. Further, the characterizing of dynamic modal activity into discrete bins
and levels of speeds and accelerations could result in a certain level of error. Current
research efforts are attempting to validate the approach.
6.4. Facility and Gridded Emissions
The spatial errors associated with the emission estimates come from
aggregation. No spatial manipulation procedures are used to generate the facility-level
emissions estimates. The facility-level estimates are. however, impacted by non-
131
-------
spatial errors. The gridded emission estimates are generated by aggregating facility-
estimates to vector grid cells of a user-defined size. During this process, spatial errors
are incurred.
6.4.1. Facility Emission Estimates
New errors introduced to the facility-level emission estimates are generated by
the process for determining emission rates. The emission modes (engine start, running
exhaust) have gram per start or gram per second rates predicted by the hierarchical
tree-based regression. The resulting emission rate values are discrete, with known
confidence bounds. The accuracy of the emission rates is affected by the size and
representativeness of the emission test dataset. The emission rates used in the model
were developed from a dataset of approximately 3000 vehicle tests using about 700
individual vehicles. Currently, improvements and additions are expanding the dataset
to over 10,000 tests.
6.4.2. Gridded Emissions
The errors associated with the aggregation of facility emissions to user-defined
grid cells are spatial in nature. Vector grid cell polygons are overlaid with the SZ, MR,
and MZ entities using the techniques described in sections 6.1.1 and 6.1.2. The 'fuzzy
tolerance' used in this process is set as low as processing time will allow. There won't
be any shared boundaries in this overlay technique and high 'fuzzy tolerance' will only
degrade the spatial quality. Once the polygons and lines are split by the grid cells, the
emission estimates are weighted (the proportion of the new entities' area or length and
the original spatial entities' area or length) and summed by grid cell.
The size of grid cell selected by the user impacts the cell's accuracy. Larger
cells will have more accurate estimates because errors (unbiased error) at the facility
levels can be offset by aggregation. Small grid cells will have fewer entities falling
within its borders, reducing the number of values to draw from. Further, grid cell sizes
falling below the spatial accuracy of the origin datasets could spatially misrepresent
the locations of emission estimates. Larger cells have the advantage of absorbing
errors related to absolute position. Figure 6.5 shows grid cell aggregations from the
sample study area described in chapter 5. Four levels are shown: 100 meter, 250
meter, 500 meter, and 1000 meter. The figure is useful in looking at the total
emissions from different levels of aggregation. While the 1000 meter (1 km) grid cell
is expected to be used for future photochemical models, additional information for
research can be gleaned from smaller cell sizes.
132
-------
100 Meter
250 Meter
500 Meter
1000 Meter
Figure 6.5 - Sample Grid Cell Aggregations
6.4.3. Sensitivity of Model
The model sensitivity can be measured in two ways: estimate accuracy and
locational accuracy. The sensitivity of the estimate accuracy can be shown by running
the model with a full range of input variables. The sensitivity of the locational
accuracy depends on the spatial allocation of the estimate, given a full range of
influential factors.
Figures 6.6 through 6.11 show how the emission rate varies for each
technology group, level of service (LOS), and road grade. The graphs are for interstate
activity only because speed and acceleration data for lower classifications do not yet
exist. The percentage of the sample area's regional fleet in each technology group is
provided in the graph as well. The very low percentage of some technology groups is
the result of the problems mentioned earlier (section 6.2) regarding the determination
of vehicle technologies.
All the technology groups have substantial estimated increases in emission
rates for LOS F. The speed and acceleration profiles for interstate LOS F show
substantially more variability in speeds and accelerations. Other LOS impacts are
133
-------
fairly static, slightly increasing as traffic flow degrades from LOS A to LOS D.
However, during LOS F activity (volume to capacities greater than 1.0 and average
speeds less than 30 MPH), the model is predicting that emissions rates increase
substantially.
The impact of road grade is seen for CO normal emitters, HC high emitters,
and NOx high emitters. The graphs indicate that these technology groups have higher
emission rates for steeper, grades. As mentioned previously, the impacts of grade may
be substantially under-predicted. Currently, the model adjusts the acceleration rates of
vehicles based on the road grade. There are no mechanisms in the model that adjust
emission rates based on engine load. It is expected that emission rates will vary
significantly once these impacts are considered.
Unlike the emission estimates, the locational sensitivity of the model is not the
result of a series of calculations. Estimates are allocated to zones or lines based on
input data conditions. For example, the return trip of a home-based-shopping (HBSH)
trip begins in a shopping area (commercial land use) and ends at home (residential
land use). If the TAZ with a HBSH attraction has commercial land use within its
boundaries, the emissions from the engine start are allocated evenly to all commercial
areas. If no commercial land use is indicated by the data, the engine start emissions
are all allocated evenly to the entire TAZ. All the sample area TAZs had residential
and non-residential land uses, and all but two had commercial land uses.
134
-------
0.0
0.6
LOS
Tech. Groups
(Number represents percent of fleet,
grade increases right to left for each
group)
Figure 6.6 - CO normal emitter technology group emission rates by LOS and
grade
135
-------
LOS
g/km
0.2
0.1
0.0
Tech. Groups
(Number represents percent of fleet,
grade increases right to left for each
group)
Figure 6.7 - CO high emitter technology group emission rates by LOS and grade
136
-------
Tech. Groups
(Number represents percent of fleet,
grade increases right to left for each
group)
Figure 6.8 - HC normal emitter technology group emission rates by LOS and
grade
137
-------
LOS
0.0
0.0
0.5
0.2
Tech. Groups
(Number represents percent of fleet,
grade increases right to left for each
group)
Figure 6.9 - HC high emitter technology group emission rates by LOS and grade
138
-------
LOS
g/km
3.1
41.4
0.0
42.2
23.5
26.4
Tech. Groups
(Number represents percent of fleet,
grade increases right to left for each
group)
Figure 6.10 - NOX normal emitter technology group emission rates by LOS and
grade
139
-------
LOS
g/km
Single Tech. Group
(1.4 percent of the fleet, grade
increases right to left)
Figure 6.11 - NOx high emitter technology group emission rates by LOS and
grade
140
-------
6.4.4. MEASURE vs. MOBILESa
To_ compare the USEPA's MOBILESa and the new HTBR emission rates used
in MEASURE, emission rates were determined for each speed and acceleration bin (0-
80 mph, -10.0 to 10.0 mph/sec). Figures 6.12 to 6.17 show these profiles. Both
emission rate models were used for each pollutant. The sample area results are also
provided, showing a comparison between the hourly total grams of each pollutant.
While much remains to be validated with MEASURE, this comparison provides some
evidence for the future development of modal emission rate models.
Both data sets used in the analysis included the regional fleet distribution for
the study area. The MOBILESa rates were the running exhaust zero mile base
emission rates (deterioration and start fractions effects removed). The HTBR rates
used in MEASURE were similar; no start emissions or deterioration effects were
included.
All the graphs show significant differences. The biggest impacts is the fact the
MOBILESa does not vary emissions by acceleration. A vehicle traveling at an average
speed of 50 mph with minor variations in acceleration and deceleration is predicted to
have the same emission rate as one with large variations. MEASURE indicates that
these variations may have significant impacts on emission rates at certain thresholds of
speed and acceleration activity.
141
-------
ie nur
Velocity (mph)
Acceleration (mph/aec)
Figure 6.12 - MEASURE g/sec CO emission rates by velocity and acceleration for
the study area's vehicle fleet
142
-------
Velocity (mph)
Acceleration (mph/sec)
00
Figure 6.13 - MOBILESa g/sec CO emission rates by velocity and acceleration for
the study area's vehicle fleet
143
-------
10
Acceleration (mph/sec)
Velocity (mph)
-------
Acceleration (mph/sec)
Velocity (mph)
Figure 6.15 - MOBILESa g/sec HC emission rates by velocity and acceleration for
the sample area's vehicle fleet
145
-------
0.03
0.025
I
0.02
0.015
0.01
0.005
8075
10
Acceleration (mph/sec)
'50 45
Velocity (mph)
-10
10 5
Figure 6.16 - MEASURE g/sec NOx emission rates by velocity and acceleration
for the sample area's vehicle fleet
146
-------
0.03
0.025
7.5
0.5
Acceleration (mph/sec)
Velocity (mph)
-10
Figure 6.17 - MOBILESa g/sec NOx emission rates by velocity and acceleration
for the sample area's vehicle fleet
147
-------
6.4.5. Conclusion
There are a variety of errors generated by model procedures. Future validation
efforts will quantify the errors so that confidence bands can be predicted for the
estimate value and position. During the process, particular attention should be paid to
the estimates of the spatial variability of-the operating fleet. Clearly, this model
component has the greatest potential for spatial error. The use of regional temporal
factors create significant non-spatial errors, particularly in off-peak hours. As long as
accurate information is fed to the model, errors resulting from modeling procedures
are significantly reduced (polygon overlay error, trip disaggregation error, etc.).
Minimum grid cell sizes should be assessed for each model scenario.
The new modal emission rates indicate that vehicle technologies and vehicle
operating profiles (speed and acceleration) have significant impacts on emission rates.
While the new emission rate models need to be validated, there is strong evidence that
MOBILE5a is insensitive to important emission-specific vehicle activity.
148
-------
7. ANTICIPATED CONTRIBUTIONS TO MOTOR VEHICLE
EMISSIONS ASSESSMENTS AND RECOMMENDED
RESEARCH
This chapter will review the contribution to the fields of transportation and air
quality planning presented in this report, and present a future research agenda.
Individuals and groups have developed spatially-resolved emission estimates
previously, but none have done so in a single, comprehensive unit. Groups have used
geographic information systems (GIS) as a pre-processor, preparing data for outside
modeling, and they have been used as a post-processor to help visualize results. The
approach described in this report relies on the capabilities of GIS to manage the
process from start to finish.
Research into transportation and air quality has made significant progress in
the last several years, but results have not been formally incorporated into
comprehensive and flexible modeling regimes. The proposed model framework
successfully incorporates existing emission research into a single model. Further, the
model is flexible to the addition of new research, an important design element needed
due to the dynamic condition of transportation and air quality research. Finally, this
report provides important insight into the current conditions of emission-specific
spatial data, and provides a tool that can be used to define the limits of disaggregate
modeling approaches. This exploration into the spatial modeling of exhaust emissions
provides substantial progress towards the development of computer tools that can aid
metropolitan transportation planners in their attempt to identify the impacts of
transportation change on air quality.
7.1. Impacts
The potential impacts of this research are significant. It has been widely
recognized that there are theoretical problems with the current modeling regime,
especially with the speed correction factor emission rate models. If a spatially
resolved modal emission model becomes accepted for use for conformity and
inventory modeling, the types of mitigation strategies available to local and state
governments change dramatically. Under the current modeling system, transportation
planners and engineers have only two ways to reduce emissions: reduce vehicle miles
of travel and/or optimize average speeds to points deemed significant by the models.
Both options probably reduce mobility and accessibility enjoyed by the transportation
149
-------
systems. If modal models are developed, much more diverse and creative strategies
become available. Any strategy that reduces the number of high-emitting vehicles or
reduces the occurrence of hard accelerations and decelerations will reduce mobile
emissions. Reducing volume may be less important than improving traffic flow
through ITS strategies, signal timing, or even lane additions. The new modal
approaches may show that mobility and accessibility can increase as mobile emissions
decrease.
Further, spatially-resolved estimates allow planners to prioritize certain
locations for mitigation strategies because of their disproportional contribution to
ozone formation due to topographic or climatic features. The spatially-resolved
estimates at proper resolutions allow local transportation planners and traffic engineers
to develop small-scale changes that reduce the net mobile emissions produced in their
jurisdictions.
One other impact is that new car standards may be altered to reduce the
occurrence of fuel enrichment resulting from high power demand. This may divert
mobile emission reduction strategies away from operational conditions and more
towards those strategies that impact engine starts.
7.2. Major Contributions
This section provides a discussion of how the objectives described in Chapter 1
were accomplished. The objectives were:
• Develop an automobile exhaust emissions model that maximizes
comprehensiveness, flexibility and user friendliness.
• Provide a research tool that allows for the testing of variable levels of motor
vehicle emission model spatial aggregation.
• Demonstrate the benefits of using GISfor emissions modeling.
• Identify research and data needs for improved spatial and temporal emissions
modeling.
7.2.1. Model Design and Development
Chapter 3 lists specific model design parameters that were identified through
background research. This section describes how the model is successful or
unsuccessful in accomplishing those design goals.
The following model design parameters were successfully included:
• All estimates (emissions, vehicle activity, etc.) must be capable of being validated.
• The model must be designed to easily incorporate new findings.
150
-------
• The model must use available, or nearly available, data.
• The model must use as large a spatial scale as data will allow.
• Develop estimates of the production of automobile exhaust pollutants CO, HC, and
NOX in space and time.
• Separate and quantify high-emitting vehicle emissions.
• Include SCF emission rates.
• Include emission rates from the statistical approach.
• Include activity measures from the travel demand forecasting models.
• Prepare for inputs from future simulation models.
• Utilize geographic information systems.
• Appropriate documentation.
• Appropriate terminology.
• Modular system design.
• Open input and output data formats.
• Intuitive model process.
• Easy to understand and use.
• Model should reside in a GIS.
The following model objectives were only partially accomplished:
• The model must produce automobile exhaust emission estimates that are capable
of being statistically verified.
The resulting model is developed around data and procedures that have
stochastic distributions. This factor makes the model statistically verifiable.
However, actual verification of the current model results must wait until individual
components are validated.
• Anthropogenic NOX estimate accuracy important in predicting ground-level ozone.
NOX is not treated in a manner distinguishing it from the other pollutants.
However, the inclusion of modal parameters allows better predictions of NOX, which
varies with speed and acceleration.
• Comprehensive representation of vehicle technologies.
As mentioned in chapter 6, substantial differences exist between fleet averages
and the ability of the MEASURE model to comprehensively represent the operating
fleet in space and time. The major problems stem from the need to have a good
routine for identifying vehicle characteristics given the vehicle identification number.
When this is done, comprehensive representation will be accomplished. On-road
vehicles distributions need more research.
• Separate start, hot-stabilized, and enrichment emission quantities and locations.
Start emissions are separate. Hot-stabilized and enrichment emissions are
combined into the running exhaust estimates. This approach includes enrichment,
151
-------
unlike other models, but not separately. Breaking the two modes (enrichment and
running exhaust) into separate procedures requires more information unavailable for
this stage of the research. However, the model framework can easily incorporate
separate procedures.
7.2.2. Tool for the Exploration of Spatial Aggregation
purpose of developing a tool that has flexible aggregation capabilities is
that gridded emissions are required to predict ambient of levels of ozone and other
pollutant concentrations. Future photochemical models will require a minimum one
kilometer grid cell aggregate estimate of mobile source pollutant production by hour.
As research into ambient air quality is conducted, the spatial scale of ozone formation
and pollutant dispersion will continually be redefined, placing spatial parameters on
input data. Further, emission rate models used for inventory purposes will have to
have a level of spatial aggregation based on data availability and algorithm accuracy.
To aid the local transportation planners in their efforts to reduce automobile pollution,
models must develop accurate estimates for transportation facilities. The facility-
estimates are aggregations of individual vehicles over time. Therefore, it is important
that the amount of aggregation or disaggregation that can be accomplished with
existing data and knowledge is identified. These issues are crucial in defining the
scope of research being conducted around the country in emission modeling.
The current model has the capability to explore levels of spatial aggregation. It
can use data from any size zonal aggregation, and it can re-allocate estimates to any
size grid cell. Once validated, it can be used to help research the issues mentioned in
the previous paragraph. The actual impacts of various levels of spatial aggregation on
the accuracy of the emission estimates will vary with the spatial quality of the input
data. As the spatial structure varies, so will the accuracy.
7.2.3. Value of Geographic Information Systems
Geographic information systems provide numerous advantages to the spatial
modeling of exhaust emissions.
• Spatial data organization
Data in the model were organized based on their spatial character. Structuring
the multiple layers of data in this manner provides data connectivity that would be
difficult without GIS and topology.
• Spatial data joining
During the modeling process, datasets of different characteristics are merged
together to form a single entity. Specifically, GIS allowed the travel demand
forecasting model network, to have improved spatial resolution by conflating to a
152
-------
spatially accurate road database. This capability of GIS also allowed linkages to occur
between the various area sources of information (TAZs, land use, Census, etc.).
• Spatial query
GIS provided the ability to search data by locational parameters. Specifically,
the technique used to predict the on-road fleet distribution required for the
identification of the fleet registered within a certain distance from the individual road
segments.
• Spatial aggregation
GIS provided the ability to aggregate irregular polygon data and line data into
regular user-defined grid cells. This capability makes GIS vital for efficiently
developing mobile emission inventories, regardless of the modeling approach used.
• Spatial data visualization
The map-making and graphic display capabilities found in most GISs are
extremely useful in communicating model results to individuals from various technical
backgrounds. Given the importance of mobile emissions in determining transportation
improvements, this feature has significant value.
7.3. Future Research
Future research is recommended in three major areas; model validation, model
improvement, and model additions. Without model validation, the ultimate value of
the model described in this report is lost. During model development and testing, it
became evident that certain strategies could be improved with designed experiments.
The following sections describe some specific actions that can be taken to expand the
model for use by transportation and air quality planners.
7.3.1. Model Validation Strategies
The model validation strategies follow the modular nature of the software.
Testing each module's results through designed experiments would answer the
accuracy questions that would allow the model to proceed beyond the prototype stage.
Prior to model validation, the input datasets should undergo significant quality
assurance testing to identify errors and temporal conflicts with other datasets.
7.3.1.1. Spatial Environment
The spatial environment module accuracy relies on the spatial accuracy of the
input data, varying with new implementation sites. Procedures used in the model to
153
-------
manipulate this information do not require validation. Each new modeled area should
undergo a data verification and validation stage. At minimum, this should include
identifying the last date the information was updated and the estimated absolute spatial
accuracy of the data.
7.3.1.2. Fleet Characteristics
The process used to develop fleet characteristics needs substantial validation
efforts. The intention of the modules is to develop an accurate profile of the operating
fleet. The problems identified with the VIN decoder and 'lookup' routines can be
solved through database and software development and are not mentioned in the
validation phase. However, their effects on the accuracy of the estimate are included
in the list of recommended validation efforts.
• Zonal fleet distribution study: A study is needed that can identify the distribution
of vehicles that are registered within a zone. This could be accomplished through
the ingress, egress study of a number of neighborhoods. The difficulty will be
defining an appropriate sample size and definition (income, family size, etc.).
• On-road fleet distribution study: A study is needed that can identify the technology
distributions of the oq-road fleet. Currently, data capable of doing this have been
collected (using video cameras) for over fifty sites. The difficulty is developing
sample sets for different road classes at different times of the day. The current
equipment can only be used across one lane of traffic, limiting the scope.
7.3.1.3. Vehicle Activity
Since vehicle activity estimates rely heavily on the travel demand modeling
process, efforts to validate the travel model significantly improve the ability of this
model to calculate errors. However, validating the engine start estimates can be
accomplished through the same ingress / egress study mentioned previously.
• Engine start activity study: A study is needed that can identify the number of
engine starts that occur by time of day within a zone. This could be accomplished
through the ingress, egress study of some neighborhoods. However, this study
would not need the video camera, but could rely on loop detectors. This would
also allow data to be collected over a long period of time.
• Road segment activity study: A study is needed that can measure the volume and
average speed of road segments modeled in the travel demand model by hour of
the day. The Atlanta Advanced Traffic Management System (ATMS) could be
used to validate interstate estimates. Other road classifications could be studied
using other techniques. Again, an appropriate sample size will have to be
determined.
• Speed and acceleration profile study: Studies that measure the speed and
acceleration profiles accuracy are needed '"Ms is currently being accomplished by
researchers at Georgia Tech for interstate ramps.
-------
7.3.1.4. Facility and Gridded Emissions
Facility emission estimates must be validated. Currently, remote sensing
technology allows for road segment, hourly pollutant production estimates to be
accomplished. The devices do not measure NOx, but it can be estimated using other
pollutant concentrations. Measurements on multi-lane roads are difficult to do. An
alternative for major roads is an upwind-downwind study where sensors are placed at
regular intervals along both sides of a road. This type of study is expensive and
unreliable in areas with large amounts of background pollution.
Outside of the remote sensing data collection, little can be done to directly
measure emission production from operating vehicles. Many researchers are working
on the issue, and technologies may develop that would make this possible.
7.3.2. Model Algorithm Improvement
There are some specific issues in the model design that could be studied to
improve the accuracy of the estimates. While model validation is important in
measuring current capabilities, these issues could improve the ability of the modules to
accurately predict their phenomena.
• Home location vs. registered address: The registered dataset includes address
fields that are supposed to represent the homes sites of the vehicles. Because
registration tax rates differ among jurisdictions, and because people move, there is
need to identify the proportion of the database that has incorrect data.
• Fraction of total vehicle operation by vehicle type: The registration dataset
represents all vehicles that are licensed to operate on the road. The actual
operating fleet may look quite different. It was evident in the two sites discussed
in section 6.2.2 that the operating fleet may be much newer than the registered
fleet.
• On-road vehicle distribution search pattern: In section 6.2.2, there was some
evidence that indicated that the radial search pattern used in the model may be
inappropriate for determining a local operating fleet. Research into the size and
shape of the search pattern could significantly improve the capability of predicting
the on-road fleet distribution.
7.3.3. Model Additions Research
The current model scope is limited to automobile exhaust emissions. Moving
to a complete mobile emissions model involves adding much more information and
data. Some of the major items are listed below:
• On and off network grade distributions and impacts: Comprehensively, the
impacts of grade on engine load have not been identified in the research. Since
road grade has spatial variability and could have significant impact on the' load on
155
-------
an engine, it should be included in the research design. This may mean moving to
a more detailed emission rate model thai has emission rates for engine load
conditions.
• More speed / acceleration matrices: The model needs more speed and acceleration
data to have a comprehensive view of modal activity on all road types. Currently,
the model is limited to eleven different profiles.
• Intersection activity: Intersections are the one facility-type missing from the
current model. Intersections will be significant to producing accurate emissions
due to the extreme variability of modal activity.
• Other motor vehicles types: Currently, only automobiles are modeled because there
are only a few vehicle emission tests for non-autos. A comprehensive mobile
source model must include all vehicles types.
• Load-based approach: A load-base approach to predicting emissions will allow
enrichment emissions to be separately identified, an original model design
objective.
• Non-exhaust mobile emissions: Exhaust emissions only make up a portion of the
overall mobile emission modes. Evaporative emissions need to be included in
future models.
• External / internal trips: Currently, external / internal trips are excluded from the
models predictions of start activity. The return trips of these vehicles are ignored,
and they could represent a significant portion.
Overall, .the model was successfully designed and developed according to
research backed parameters. Substantial progress towards the development of a
comprehensive mobile source inventory / impact model has been accomplished.
156
-------
8. REFERENCES
[Awuah-Baffour 1997] Robert Awuah-Baffour, Wayne Sarasua, Karen Dixon,
William Bachman, and Randall Guensler; "GPS with an Attitude";
Transportation Research Record: Transportation Research Board, Washington DC,
1998.
[Bachman 1998] Bachman, William, Jessica Granell, Randall Guensler, and John
Leonard; "Research Needs Jji Determining Spatially-Resolved Sub-fleet
Characteristics"; Transportation Research Record; Transportation Research Board,
Washington, DC, 1998.
[Bachman 1996] Bachman, William, Wayne A. Sarasua, and Randall Guensler; "A
GIS Framework for Mobile Source Emissions Modeling"; Transportation
Research Record No. 1551. National Academy Press; January, 1996.
[Earth 1996] Barth, Matthew, Feng An, Joseph Norbeck, and Marc Ross; "Modal
Emissions Modeling: A Physical Approach;" Transportation Research Record;
National Academy Press; January, 1996.
[Benson 1989] Benson, Paul; "CALINE4 A Dispersion Model for Predicting-
Pollutant Concentrations Near Roadways" (FHWA/CA/TL-84/15); State of
California Department of Transportation, Division of New Technology and
Research; Sacramento, CA; November, 1984, Revised June, 1989.
[Bruckman 1991] Bruckman, Leonard, Ronald J. Dickson, and James G. Wilkinson;
"The Use of GIS Software in the Development of Emissions Inventories and
Emissions Modeling;" Proceedings of the Annual Meeting and Exhibition of the
Air and Waste Management Association; June, 1992.
[Calspan 1973] Calspan Corporation; "Automobile Exhaust Emission Surveillance";
Environmental Protection Agency #APTD-1544; NTIS PB220-775; Office of
Mobile Source Air Pollution Control; Ann Arbor, MI; May, 1973.
[CARB 1990] California Air Resources Board; Emission Inventory 1987; Technical
Support Division; Sacramento, CA; March, 1990.
[Carlock 1993] Carlock, Mark; "An Analysis of High Emitting Vehicles in the On-
Road Motor Vehicle Fleet"; The Emission Inventory: Perception and Reality,
Proceedings of an International Specialty Conference; Air and Waste Management
• Association; Pittsburgh, PA; October, 1993; pp 207-228.
157
-------
[Grant 1996] Grant, Chris, Randall Guensler, and Michael D. Meyer; "Variability of
Heavy-Duty Vehicle Operating Mode Frequencies for Prediction of Mobile
Emissions"; Proceedings of the 89th Annual Meeting; Air and Waste Management
Association; Pittsburgh, PA; June, 1996.
[Groblicki 1990] Groblicki, Peter J.; Presentation at the California Air Resources
Board Public Meeting on the Emission Inventory Process; General Motors
Re .earch Laboratories; Warren, MI; November 5, 1990.
[Guensler 1993a] Guensler, Randall; "Data Needs for Evolving Motor Vehicle
Emission Modeling Approaches"; Transportation Planning and Air Quality II; Paul
Benson, Ed.; American Society of Civil Engineers; New York, NY; 1993.
[Guensler 1993b] Guensler, Randall; "Vehicle Emission Rates and Average Vehicle
Operating Speeds"; Dissertation; submitted in partial satisfaction of the
requirements for the degree of Doctor of Philosophy in Civil/Transportation
Engineering; Advisor: Daniel Sperling; Department of Civil and Environmental
Engineering, University of California, Davis; Davis, CA; December, 1993.
[Guensler 1994] Guensler, Randall; "Vehicle Emission Rates and Average Vehicle
Operating Speeds"; Institute of Transportation Studies, University of California,
Davis, CA; 1994.
[Hallmark 1996] Hallmark, Shauna and Wende O'Neill; "Integrating GIS-T and Air
Quality Models for Microscale Analysis"; Transportation Research Record No.
1551; National Academy Press; 1996.
[Harvey 1991] Harvey, G. and E. Deakin; "Toward Improved Regional
Transportation Modeling Practice"; National Association of Regional Councils;
Washington DC; December, 1991.
[Heywood 1988] Heywood, John B.; Internal Combustion Engine Fundamentals;
M-Graw-Hill Publishing Company; New York, NY; 1988.
[Jacobs 1990] Jacobs, Paul, Donald J. Churnich, and Mark A. Burnitzki; "Motor
Vehicle Emissions and their Controls"; California Air Resources Board;
Sacramento, CA; July, 1990.
[Johnson 1993] Johnson, Elmer; "Avoiding the Collision of Cities and Cars, Urban
Transportation Policy for the Twenty-first Century"; Report by The Academy of
Arts and Sciences and The Aspen Institute; Chicago, IL; September, 1993.
[Joy 1992] Joy, Richard W.; "Drive Thru Windows - A Case Study"; Proceedings of
the Transportation Modeling Tips and Trip-Ups Conference; Air and Waste
Management Association; Pittsburgh, PA; March 11-12, 1992.
158
-------
[Kunselman 1974] Kunselman, P., H.T Me Adams, C.J. Domke, and M.E. Williams;
"Automobile Exhaust Emission Modal Analysis Model"; Calspan Corporation;
' Buffalo, NY; Environmental Protection Agency (EPA-460/3-74-005); Office of
Mobile Source Air Pollution Control; Arin Arbor, MI; January, 1974.
[Meyer 1997] Meyer, Michael D.; "A Toolbox for Alleviating Traffic Congestion and
Enhancing Mobility"; Institute of Transportation Engineers for the Federal
Highway Administration; Washington DC; July, 1997.
[Miller 1995] Miller, Terry L., Arun Chatterjee, and Cheng Ching; "Travel Related
Inputs to Air Quality Models: An Analysis of Emissions Model Sensitivity and the
Accuracy of Estimation Procedures"; Transportation Congress; Volume 1, pg
1149; 1995.
[NCHRP 1993] NCHRP Report 359; "Adaptation of Geographic Information
Systems for Transportation"; Transportation Research Board; National Academy
Press; Washington, DC; 1993.
[NRC 1991] National Research Council, Committee on Tropospheric Ozone
Formation and Measurement; Rethinking the Ozone Problem in Urban and
Regional Air Pollution; National Academy Press; Washington DC; 1991.
[Orthofer 1995] Orthofer Rudolph and Wolfgang Loibl; "GIS-Aided Spatial
Disaggregation of Emission Inventories"; The Emission Inventory: Programs and
Progress; Presentation at the Air and Waste Management Specialty Conference;
RTF, NC; October, 1995.
[Outwater 1994] Outwater, Maren and William Loudon; "Travel Forecasting
Guidelines for the Federal and California Clean Air Act"; Transportation Research
Record; Transportation Research Board; January, 1994.
[Pollack 1992] Pollack, Alison, Jeremy G. Heiken, and Robert A. Gorse;
"Comparison of Remote Sensing Data and Vehicle Emission Models: The High
Proportion of Emissions from High Emitting Vehicles"; Proceedings of the 85th
Annual Meeting of the Air and Waste Management Association; Pittsburgh, PA;
June, 1992.
[Pozniak 1980] Pozniak, Donald J.; "The Exhaust Emission and Fuel Characteristics
of an Engine During Warmup, A Vehicle Study (800396)"; Society of Automotive
Engineers; Warrendale, PA; February, 1980.
[Reingruber 1994] Reingruber, Michael and William Gregory; The Data Modeling
Handbook; John Wiley & Sons, Inc.; New York, NY; 1994.
159
-------
[Sabate 1994] .Sabate, Shelley and Archana Agrawal; "Proposed Methodology for
Calculating and Redefining Cold and Hot Start Emissions"; California Air
Resources Board; 1994.
[Siwek 1997] Siwek, Sarah J.; Summary of Proceedings, EPA-FHWA Modeling
Workshop; Ann Arbor, MI; 1997.
[SOS 1994] Southern Oxidants Study; "The State of the Southern Oxidants Study:
Policy-Relevant Findings in Ozone Pollution Research, 1988-1994"; W.L.
Chameides and E.B. Cowling, eds.; North Carolina State University; Raleigh, NC;
1994.
[Souleyrette 1991] Souleyrette, Reginald, Shashi Sathisan, David James, and Soon-tin
Lim; "GIS for Transportation and Air Quality Analysis"; Proceedings of the ASCE
Urban Transportation Division National Specialty Conference on Transportation
Planning and Air Quality; Santa Barbara, CA; July, 1991.
[Stopher 1993] Stopher, Peter; "Deficiencies of Travel-Forecasting Methods Relative
to Mobile Emissions"; Journal of Transportation Engineering, ASCE; Vol. 119,
No. 5; September/October, 1993.
[Tomeh 1996] Tomeh, Osama; "Spatial and Temporal Characterization of the Vehicle
Fleet as a Function of Local and Regional Registration Mix: Methodological
Development"; Dissertation, School of Civil and Environmental Engineering,
Georgia Institute of Technology; Atlanta, GA; May, 1996.
[USEPA 1995] US Environmental Protection Agency; "National Air Pollutant
Emission Trends, 1900-1994"; (EPA-454/R-95-011; NTIS PB 96-135678); Office
of Air Quality Planning and Standards; Research Triangle Park, NC; October,
1995.
[USEPA 1997] US Environmental Protection Agency; Congressional Subcommittee
Testimony by Carol Browner; CSPAN; Washington, DC; 1997.
[Venigalla 1995a] Venigalla, Mohan, Terry Miller, and Arun Chatterjee; "Alternative
Operating Mode Fractions to Federal Test Procedure Mode Mix for Mobile Source
Emissions Modeling"; Transportation Research Record No. 1472. p. 35-44;
National Academy Press; Washington, DC; 1995.
[Venigalla 1995b] Venigalla, Mohan, Terry Miller, and Arun Chatterjee; "Start
Modes of Trips for Mobile Source Emissions Modeling"; Transportation Research
Record No. 1472. p. 26-34; National Academy Press; Washington, DC; 1995.
[Washington 1994] Washington, Simon; "Estimation of a Vehicular" Carbon
Monoxide Modal Emission Model and Assessment of an Intelligent Transportation
160
-------
Technology"; Dissertation; Department of Civil Engineering, University of
California, Davis; Davis, CA; 1994.
[Washington 1995] Washington, Simon; "Modal Activity Correction Factors for
Implementing Emission Rate Estimates"; Transportation Research Record;
Transportation Research Board; January, 1995.
[Washington 1996] Washington, Simon; "Considerations for the Development of
New Mobile Source Emission Models"; Proceedings of the 76th Annual
Transportation Research Board; Washington DC; January, 1996.
[Wolf 1998] Wolf, Jean, Simon Washington, Randall Guensler, and William
Bachman; "High Emitting Vehicle Characteristics Using Regression Tree
Analysis"; Transportation Research Record; Transportation Research Board,
Washington DC; 1998.
[Worboys 1995] Worboys, Michael F.; GIS: A Computing Perspective; Taylor and
Francis; London, UK; 1995.
161
-------
APPENDIX A
DATA DICTIONARY
The data dictionary will explain the design and function of each file and its
underlying fields / attributes. It will divide the data into two components: input files
and output files. Further, all files are ARC/INFO coverage files, Dbase 4 files, or
ASCII text files. Chapter 4 provides a functional overview of all the input and output
files. ARC/INFO coverages consist of several related files that are maintained by the
software. For these databases, only the attribute file is described because it is usually
the only one edited by the user. Fields in the attribute files maintained by ARC/INFO
(area, perimeter, length, cover#, and cover-id) are not profiled.
The directory structure consists of 16 directories. All directory contents are
discussed in detail later:
1. ami - Contains 'ami' scripts. These ASCII files are ARC/INFO macros that
are called throughout the process.
2. Code - Contains 'C' programming code. The ASCII code files, and their
binary object and executable files are stored in this directory.
3. Em - Contains emission estimate results. All final emission estimates are
written to this directory.
4. Grade - Contains all files required to estimate the road grade for the study
area.
5. Grid - Contains all vector grid coverages that are desired for the
assessment (e.g., 1 km, 2km, 4km).
6. Landmarks - Contains all point data for specific landmarks (schools,
universities, etc.)
7. Lookup - Contains all ASCII look-up files. These consist of the SCF
emission rate file and Vehicle Test Weight file.
162
-------
8. Modalmats - All the pre-processed modal matrices required for the
aggregate-modal emission rates are stored here.
9. Raster - All the gridded, hourly, raster outputs are located here.
10. Raw - Contains readable copies of all the input files.
11. Road - Contains all data related to the road network.
12. Sa - Contains all the ASCII speed / acceleration matrices used to calculate
average speed and modal conditions.
13. Temp - Initially contains nothing. But during processing, it is a working
directory where temporary files are written and erased.
14. Templates - Contains empty data templates that speed up the writing of
files with many fields. These are all INFO files.
15. Tg - Contains all vehicle-related information needed to determine
emission-specific technology group distributions.
16. Zone - Contains all zonal based data (landuse, census data, TAZs, etc.).
INPUT Files:
ARC/INFO Coverages:
1. GRID - GRID is a vector polygon coverage that will be used as the basis for the
output raster database. This database should consist of regular square sized
polygons of any user-defined dimension. All attribute data are found in the
polygon attribute table call 'grid.pat' One way to get this database for the study
area is to use the ARC/INFO command 'generate' with the 'fishnet' option. This
database is stored in the 'grid' directory.
COLUMN
1
5
9
13
ITEM NAME
AREA
PERIMETER
GRID#
GRID-ID
WIDTH OUTPUT TYPE N.DEC
17 GDID
4
4
4
4
4
12
12
5
5
5
F
F
B
B
B
3
3
-
-
-
The field GDID represents a unique ID field that is initially set to the internal id
GRID# + 1. One is added because the first record is an external polygon
representing undefined space. This undefined space is therefore assigned a 0 id.
The GDID is used to track disaggregation that occurs during processing.
163
-------
2. LANDUSE - The LANDUSE coverage is a vector polygon database consisting of
residential, commercial, and other landuses. It is used to disaggregate TAZ trip
information into smaller areas. For example, it is assumed that vehicle trips that
begin at home and end up at work, begin in residential and end in non-residential
land uses. This disaggregation allows the engine start emissions to be better
spatially defined.
4.
COLUMN ITEM NAME
1
5
9
13
AREA
PERIMETER
LANDUSE*
LANDUSE-ID
17 LU
4
4
4
4
3
'PUT
12
12
5
5
3
TYPE N
F
F
B
B
C
.DEC
3
3
-
-
-
The LU field consists of three characters describing the land use type. Acceptable
values for this first version are 'RES', 'COM', or 'UNK'. 'UNK' polygons have
landuses that are unknown or non-residential or non-commercial. For example, an
institutional land use would be designated 'UNK'.
CENSUS - The CENSUS dataset is a vector polygon coverage of socioeconomic
data. It is expected that these polygons would be Census Blocks defined in the
1990 US Census and stored in the 1994 TIGER files. Actually, they can represent
other datasets (parcel level data, local economic zones, etc.). It is important that the
fields be present as described below.
WIDTH OUTPUT TYPE N.DEC
4
4
4
4
4
8
8
8
8
12
12
5
5
5
8
• 16
10
10
F
F
B
B
B
I
F
F
F
3
3
-
-
-
-
5
4
4
COLUMN ITEM NAME
1 AREA
5 PERIMETER
9 CENSUS*
13 CENSUS-ID
17 CBID
21 HU90
25 HU90/KM
33 SOV
41 CARPOOL
The CBID field is the identifier that is tracked throughout the modeling process.
This field becomes a key field to link zonal activity and emission estimates. The
HU90 field contains the number of housing units (1990) found within the polygon.
This field is used to disaggregate home-based trips from the TAZ level to the
Census Block level. The HU90/KM is the 1990 housing units per square kilometer.
The SOV field is the fraction of 1990 workers that drove to work alone. The
CARPOOL field is the fraction of 1990 workers that carpooled to work. The
remaining trips were non-auto trips.
TAZ - The TAZ dataset is a vector polygon coverage of the local planning
organization's traffic analysis zones. The zones are used in planning agencies as a
• spatial unit, summarizing trip origins and destinations that occur within each. The
164
-------
trip estimates are used to predict the number of engine starts that occur within the
zone.
COLUMN
1
5
9
13
17
21
29
37
45
53
61
69
77
85
93
101
110
ITEM NAME
AREA
PERIMETER
TAZ#
TAZ-ID
TZID
HBW_PRD
HBSH_PRD
HBGS_PRD
HBU_PRD
HBO_PRD
NHB_PRD
HBW_ATT
HBSH_ATT
HBGS_ATT
HBU_ATT
HBO_ATT
NHB ATT
WIDTH OUTPUT TYPE N.DEC
4
4
4
4
4
8
8
8
8
8
8
8
8
8
8
8
8
12
12
5
5
5
10
10
10
10
10
10
10
10
10
10
10
10
F
F
B
B
B
f
f
f
f
f
f
f
f
f
f
f
f
3
3
-
-
-
0
0
0
0
0
0
0
0
0
0
0
0
The TZID field represents the TAZ identifier that is tracked throughout the
modeling procedures. It is a key field that is used often to link related data and
subsequent estimates. The remaining fields identify trip types that are defined by
the local travel demand forecasting models developed and used by local
transportation planners. HBW_PRD are 24-hour home-based work productions
(trips between home and work or work and home). HBSH_PRD are home-based
shopping trip productions. HBGS_PRD are home-based grade-school productions.
HBU_PRD are home-based university trip productions. HBO_PRD are home-
based other productions (trips that begin or end at home and go to or return from
someplace other than work, shopping areas, grade schools, or universities).
NHB_PRD are trips that begin and end someplace besides home. All the remaining
fields ending in ATT describe the attractions of each trip type.
ZIP code - The ZIPCODE dataset is a vector polygon database that represents 5
digit ZIP codes in the study area. The primary purpose of the ZIP code database is
to identify vehicle type distribution locations for those vehicles whose address was
unmatched. During the assessment of vehicle registration data (discussed later),
individual vehicles are assigned coordinates based on their address. When the
address location can not be successfully or confidently identified, the vehicle's
location parameter becomes its registered ZIP code. This polygon database
becomes the means of identifying location. It also provides another polygonal form
for aggregating results for comparison between ZIP codes in a region.
COLUMN ITEM NAME
1 AREA
5 PERIMETER
9 ZIPCODEtt
WIDTH OUTPUT
4 12
4 12
4 5
TYPE N.DEC
F 3
F 3
B
165
-------
13 ZIPCODE-ID 4 5 B
17 ZIPCODE 551
22 ZPID 4 5 B
The ZIPCODE field holds the 5-digit ZIP code number. The ZPID field is an ID
that is tracked throughout the model.
6. Allroads - The ALLROADS dataset is a vector line database of all roads in the
study area. The lines are used to identify the locations of emissions that occur as a
vehicle moves through the road network. Emission estimates are estimated on a
line-by-line basis. The fields are;
COLUMN ITEM NAME WIDTH OUTPUT TYPE N.DEC
1 FNODE# 4 5 B
5 TNODE# ' 4 5 B -
9 ,LPOLY# 4 5 B
13 RPOLY* 4 5 B
17 LENGTH 8 18 F 5
25 ALLROADS* 4 5 B
29 ALLROADS-ID 4 5 B
33 ARID 881
41 TFID 4 5 B
The ARID field is a unique identifier for every line. This identifier is tracked
throughout the modeling process and used a key field linking a number of related
files. The TFID field is an identifier linking the travel demand forecasting network
link. Every line in the local planning organization's network must be represented
in the ALLROADS road dataset. TFID becomes a key field for linking travel
demand forecasting model data and emission outputs.
7. Landmarks - The LANDMARKS dataset is a vector point database of schools
and universities. These landmarks are special trip generators and attractors used in
developing estimates travel behavior by-local planners. Only grade school and
university locations are used in this version.
COLUMN ITEM NAME WIDTH OUTPUT TYPE N.DEC
1 AREA 4 12 F 3
5 PERIMETER 4 12 F 3
9 LANDMARKS* 4 5 B
13 LANDMARKS-ID 4 5 B
17 CODE 4 4 C
The CODE field identifies whether the landmark is a grade school or a university.
A code of 'G09' indicates a grade school. A code of 'D43' represents a university.
The codes correspond to database definitions used by a landmark database
available in many cities.
INFO Files:
166
-------
TDFN.DAT - The TDFN.DAT datafile contains fields that are used in the travel
demand forecasting software called TRANPLAN. This file holds the predicted
road volumes, average speeds, and capacities that are output from the software.
These items area used in MEASURE to identify the number of vehicles and levels
of congestion for each modeled road segment.
COLUMN ITEM NAME
1 TFID
5 FN
11 TN
17 ASSIGN_GR
18 DIST
22 AB_OPTION
25 ABSPEED1
29 ABSPEED2
33 ABDIR
35 ABLG1
37 ABLG2
39 ABLG3
41 ABCAP
47 ABVOL
53 BA_OPTION
56 BASPEED1
60 BASPEED2
64 BADIR
66 BALG1
68 BALG2
7 0 BALG3
72 BACAP
7 8 BAVOL
WIDTH OUTPUT TYPE N.DEC
4
6
6
1
4
3
4
4
2
2
2
2
6
6
3
4
4
2
2
2
2
6
6
5
6
6
1
4
3
4
4
2
2
2
2
6
6
3
4
4
2
2
2
2
6
6
B
I
I
I
I
C
I
I
I
I
I
I
I
I
C
I
I
I
I
I
I
I
I
Only seven fields in this file are used by the model: TFID, ABSPEED1, ABCAP,
ABVOL, BASPEED1, BACAP, and BAVOL. Fields with AB refer to the travel
lanes moving in the 'from node' - 'to node' direction while BA refers to the
reverse direction. The TFID field is a key field that is used to link to the
ALLROADS attribute table. ABSPEED1 and BASPEED1 are the average
modeled speeds in hundredths of a mile per hour. ABCAP and BACAP are the
estimated capacities of the road by direction. ABVOL and BAVOL are the
estimated 24-hour volumes of the road by direction. The remaining fields are not
used by the model, but are maintained for future query capability. FN and TN are
from and to node identifiers. ASSIGN_GR is the road classification. DIST is the
actual distance of the TRANPLAN link. AB_OPTION and BA_OPTION are flags
indicated road characteristics. ABSPEED2 and BASPEED2 are usually not used,
but can represent average speed for a different set of conditions. ABDIR and
BADIR identify the direction of each lane group. ABLG1-3 and BALG1-3 are
fields that can identify factors for specific lanes or lane groups.
167
-------
2. TEMPORAL.FACTORS - The TEMPGF ^.FACTORS (in the 'templates'
directory) database contains multipliers for jarly travel activity. The file contains
one record for each hour of a day, and fields for each trip purpose and one for on-
road activity. For example, record 8 (representing 7-8 am) holds a value of
0.18080, meaning that 18.08% of daily on-road travel occurs during this period.
These data were developed for Atlanta based on regional reports on travel
behavior.
ASCII Files:
1. SCF.CSV - The SCF.CSV comma-delimited file is a lookup table for the running
exhaust gram/second emission rates from MOBILESa listed by 10-MPH
increments and model year. It was created by running MOBILESa for the given
condition (100% fleet with certain model year, average speed of certain increment)
with cold start percentages set to zero. Cold starts are calculated separately. There
are five columns of data;
- The first column is average speed in miles per hour.
The second column is the model year (1970-1994).
The third column is the CO emission rate.
The fourth is the HC emission rate.
The fifth is the NOx emission rate.
This file replaces the need to run the MOBILESa model repeatedly during
MEASURE run time.
2. ZONE.TWT and ZIP.TWT - The ZONE.TWT and ZIP.TWT comma-delimited
files identify vehicle characteristics by location. These files are the products of a
series of programs that exist outside of MEASURE. The programs process
department of motor vehicle registration data by address-geocoding and VIN
decoding all of the records. During the address-geocoding process, individual
vehicles are either successfully or unsuccessfully geocoded. Successful records are
assigned a zone identifier that is the equivalent of CBID discussed previously and
written to ZONE.TWT file. Unsuccessful records default to the ZIP code and are
written to the ZIP.TWT file. The files are identical in structure; 10 columns of
data:
The first column contains the zonal or ZIP code identifier.
The second column contains the vehicle identification number, or VIN.
The third column contains the model year.
The fourth column contains the er vssion control equipment type (4 =
Oxidation and Catalyst, 3 = Catalyst only, 2 - 'xidation only, 1 = none).
The .rifth column contains the fu-: avery type flag (4 = DS, 3 =
Throttle Body. _ = Carburetor, 1 = Anything ).
The sixth column contains the ct :ich displacement of the engine.
The seventh column contains the '. iicle test weight in pounds.
The eighth column contains a flag :>? being a CO high emitter or not.
168
-------
The ninth column contains a flag for being a HC high emitter or not.
The tenth column contains a flag for being a NOx high emitter or not.
3. SZZP.ASC - The SZZP. ASC comma-delimited file holds factors for joining
technology? group fractions from the ZIP code level to the zonal level (census
blocks). Area and address-matching rates are used to factor the ZIP code fractions
by multiplying both distributions by their appropriate match or failure rate. The
ZIP code distribution is further factored by the area of the zone divided by the area
of the ZIP code. The distributions are then combined to give the best estimate of
the zonal technology group distribution. The file has five columns of data;
The first column contains the zone identifier.
The second column contains the zone area in square feet.
The third column contains the ZIP code identifier.
The fourth column contains the ZIP code area in square feet.
The fifth column contains the address-geocoding success rate.
4. GRADE.XY - The GRADE.XY file is a comma-delimited file of points along
roads that have grade data. The file is usually developed from the output of a
dynamic attitude GPS device. The first column is the unique point identifier, the
second column is the x coordinate (using whatever projection system is standard to
the rest of the data) and the third column is the y coordinate. These locations are
matched with other data and assigned to road segments based on their location.
5. GRADE.GR - The GRADE.GR file accompanies the previous file by providing
the road grade reading at that location. The first column of this comma-delimited
file is the unique identifier, and the second is the road grade reading.
OUTPUT FILES
The output files consist of a number of intermediate and final Dbase 4 files,
ARC/INFO coverages, and raster GRID databases. Many of the intermediate outputs
are saved for comparison, making all data available for analysis. Many files are
created and processed in during individual module runs. These files are temporarily
stored until the particular module is complete. The temporary files are not discussed in
this data dictionary, only the outputs from each module.
Zonal Environment Module:
1. Sz - The SZ (start zones) database is an ARC/INFO polygon coverage that is the
spatial intersection of the input polygons CENSUS, TAZ, LANDUSE, and
ZIPCODE. Every SZ polygon has identifiers to the original polygon structures.
. .The SZ database is used as the spatial aggregation for estimating engine start
169
-------
emissions. At the end of the model, each individual polygon has an estimate nf the
amount of CO, HC and NOx emissions that result from engine starts.
COLUMN
ITEM NAME
1 AREA
5 PERIMETER
9 SZ#
13 SZ-ID
17 CBID
21 HU90/KM
29 TZID
33 ZPID
37 SZID
WIDTH OUTPUT TYPE N.DEC
3
3
4
4
4
4
4
8
4
4
4
12
12
5
5
5
16
5
5
5
F
F
B
B
B
F
B
B
B
The CBID field identifies the CENSUS polygon. The TZID field identifies the
TAZ polygon. The ZPID identifies the ZIP code polygon. The SZID field
represents a unique identifier for each 'start zone'. The HU90/KM field is the 1990
Census households per square kilometer for that polygon.
Sz.dat - The SZ.DAT info file maintains important data for each 'start zone'. It is
created by stripping attributes from the combined coverages that make up the 'start
zones'.
COLUMN
1
5
9
12
16
24
32
40
48
56
64
72
80
88
96
104
112
120
128
136
141
145
149
157
165
ITEM NAME
SZID
AREA
LU
HU90
HU90/KM
SOV
CARPOOL
HBW_PRD
HBSH_PRD
HBGS_PRD
HBU_PRD
HBO_PRD
NHB_PRD
HBW_ATT
HBSH_ATT
HBGS_ATT
HBU_ATT
HBO_ATT
NHB_ATT
ZIPCODE
TZID
CBID
RES
NONRES
COM
WIDTH OUTPUT TYPE N.DEC
4
4
3
4
8
8
8
8
8
8
8
8
8
8
8
8
8
8
8
5
4
4
8
8
8
5
12
3
5
16
10
10
10
10
10
10
10
10
10
10
10
10
10
10
5
5
5
10
10
10
B
F
C
B
F
F
F
F
f
f
f
f
f
f
f
f
f
f
f
I
B
B
F
F
F
-
3
-
-
5
4
4
0
0
0
0
0
0
0
0
0
0
0
0
-
-
-
0
0
0
170
-------
The SZID field is the unique key field that links to the SZ polygons. AREA is the area
of the SZ polygon in the current projection units. LU is the land use read from the
LANDUSE polygons. HU90 is the estimated 1990 housing units for the SZ polygon.
HU90/KM is the housing unit density in square kilometers. SOV and CARPOOL
contain the fraction of travel to work by each type. The next 12 fields, starting with
HBW_PRD, contain the estimates of the number of trips produced or attracted to the
SZ polygon for each trip type. The ZIPCODE field holds the SZ polygons 5-digit ZIP
code. TZID is the identifier to a TAZ polygon, and CBED is the identifier to the
CENSUS polygon. RES contains the area of residential land use in square kilometers.
NONRES contains the area of non-residential land use in square kilometers. COM
contains the area of commercial land use in square kilometers.
Road Environment Module:
1. Mr - The MR database is a vector line dataset of major road centerlines in the
study area. This database contains the entities that are used to aggregate emissions
that occur from running exhaust on major roads. Major roads are defined as those
modeled by the local planning organization's regional travel demand forecasting
model. These roads are considered separately because of the prognostic data
available through the forecast models. Individual lines represent road segments
that start and end at crossings of other major roads.
COLUMN
1
5
9
13
17
25
29
33
41
ITEM NAME
FNODE#
TNODEtt
LPOLYtt
RPOLY#
LENGTH
MR#
MR- ID
ARID
TFID
WIDTH OUTPUT TYPE N.DEC
4
4
4
4
8
4
4
8
4
5
5
5
5
18
5
5
8
5
B
B
B
B
F
B
B
I
B
Outside normal system fields managed by ARC/INFO, each line has two identifier
fields. The first is ARID. ARID is a key field link to the ALLROADS input
database. The second, TFID is a key field link to the TDFN.DAT input file.
Mz - The MZ database is a polygon database of minor road aggregations. These
polygons are bounded by lines from the MR database. The polygons are used to
aggregate running exhaust emissions that occur of the major road network (those
roads not modeled by regional travel demand forecasting models).
COLUMN
1
9
17
ITEM NAME
AREA
PERIMETER
MZ#
WIDTH OUTPUT TYPE N.DEC
21 MZ-ID
25 MZID
18
18
5
5
5
F
F
B
B
B
171
-------
The field MZID is a unique identifier that is used as a key field linking activity and
emissions data estimated by the model.
Engine Start Activity Module:
1. Sz-act.dbf - The SZ-ACT.DBF file is a Dbase 4 file of engine start activity
occurring within the SZ (start zone) polygons. The file is created by spatially
disaggregating regional travel information and temporal factors from the input
files.
COLUMN ITEM NAME
1 SZID
5 ESI
I
ES24
WIDTH OUTPUT
4 5
4 5
TYPE N.DEC
B
B
B
The SZID field is an identifier to an SZ polygon. The ES1-ES24 fields contains the
number of engine starts that occur within each ES polygon during that particular
hour (ES 1 = engine start from midnight to 1AM).
Running Exhaust Activity Module:
1. Mr-act.dbf - The MR-ACT.DBF file is a Dbase 4 file of running exhaust activity
parameters that occur on each MR (major road) line.
WIDTH OUTPUT TYPE N.DEC
COLUMN
1
9
17
25
33
41
49
.57
65
73
81
89
97
289
481
505
ITEM NAME
ARID
TFID
FREQUENCY
SUM_SA1
SUM_SA2
SUM_SA3
SUM_SA4
SUM_SA5
LENGTH
CLASS
CAPACITY
LANES
SPD1-24
VV.,1-24
LOJ1-24
VCR1-24
9
11
11
25
25
25
25 :
25
24
2
7
3
16
7
1
12
F
F
F
F
F
F
F
F
F
F
F
F
F
F
C
F
0
0
0
6
6
6
6
6
5
0
0
0
5
0
-
1
The field ARID is a ,
the TDFT- \T file
record (m. r: road s
less than -...^45. SI
0.045 and -0.015. S
0.015 and+0.015. 5
+0.015 ar * +0.045.,
to the ALLROADS database. The field TFID is a link to
7,QUENCY is the number of grade data points for each
nt). SUM_SA1 is the number of data points with grades
'•2 is the number of data points with grades between
,A3 is the number of data points with grades between
_SA4 is the number of data points with grades between
,M_SA5 is the number of data points with grades greater
172
-------
than +0.045. LENGTH is the length of the road segment in the current map units
(meters). CLASS is the road classification. LANES is the number of one-way
travel lanes for the road segment. SPD1-24 is the average travel speed for each
hour in a day. VOL 1-24 is the traffic volume for each hour. LOS 1-24 is the level
of service for each hour. VCR1-24 is the volume to capacity ratio.
2. Mz-act.dbf - The MZ-ACT.DBF file is a Dbase 4 file of the minor zone vehicle
activity. The file stores activity information that is used in predicting emissions.
The MZID field is an identifier linking the data to the MZID polygons. The
MEAN_TRAVE field is the aggregate mean travel time from the centroid of each
SZ polygon to the closest major road. The fields SUM_ESl-24 contain the
aggregate number of engine starts (trips) that occur within the minor zone during
the given hour.
Engine Start Technology Group Module:
1. Estg.dbf - The ESTG.DBF file contains all the emission-specific technology
group distributions for each CENSUS polygon (1990 Census Blocks). The CBID
field links to the CBID field in the CENSUS and SZ polygons. The FREQ field
shows the total number of active vehicles estimated to reside in the polygon. The
remaining fields are the distributions. Every individual vehicle is assigned a
technology group for CO, HC, and NOx. Within each pollutant, vehicles are
divided into high and normal emitters based on technology characteristics.
Therefore, the fields are as follows:
ESCON: Engine Start, CO, Normal
ESCOH: Engine Start, CO, High
ESHCN: Engine Start, HC, Normal
ESHCH: Engine Start, HC, High
ESNON: Engine Start, NOx, Normal
ESNOH: Engine Start, NOx, High
2. Esreg.dbf - The ESREG.DBF file contains all the engine start emission-specific
technology group distributions for the regional fleet. The file only contains the
group fields and a frequency field.
Running Exhaust Technology Group Module:
1. Mrtg.dbf - This file contains aggregate modal running exhaust technology group
distributions for each road segment in the MR line database. The ARID field links
to the MR.AAT and the ALLROADS.AAT attribute tables. The FREQ field shows
the total number of vehicles operating on the line in a 24-hour period. The
technology group fields are as follows:
HSCON: Running Exhaust, CO, Normal
173
-------
HSCOH: Running Exhaust, CO, High
HSHCN: Running Exhaust, HC, Normal
HSHCH: Running Exhaust, HC, High
HSNON: Running Exhaust, NOx, Normal
HSNOH: Running Exhaust, NOx, High
2. Scftg.dbf - The SCFTG.DBF file is a Dbase 4 file of the Speed Correction Factor
Technology Groups. These are the model year distributions used by MOBILE5a to
determine appropriate running exhaust emission rates. The file links to the MR and
ALLROADS line databases through the ARID identifier field. The FREQ field
identifies the number of vehicles predicted to operate on that road segment during
a 24-hour period. The remaining fields MY70-MY94 show the fraction of the
operating vehicles in each model year.
3. Rereg.dbf - The REREG.DBF file contains all the running exhaust emission-
specific technology group distributions for the regional fleet. The file only contains
the group fields and a frequency field.
Engine Start Emissions Module:
1. Es-em.dbf - The ES-EM.DBF file is Dbase 4 file of the engine start emission
estimates. The file contains the SZID identifier linking the estimate to the SZ
polygon database. The remaining fields show the hourly estimates of each
pollutant in grams. The fields are listed as CO 1-24, HC1-24, and NOx 1-24.
Running Exhaust Emissions Module:
1. Mr-em.dbf - The MR-EM.DBF file is Dbase 4 file of the aggregate modal running
exhaust emission estimates. The file contains the ARID identifier linking the
estimate to the MR and ALLROADS line database. The remaining fields show the
hourly estimates of each pollutant in grams. The fields are listed as CO 1-24, HC1-
24, and NOx 1-24.
2. Mz-em.dbf - The MZ-EM.DBF file is Dbase 4 file of the minor road emission
estimates. The file contains the MZID identifier linking the estimate to the MZ
polygon database. The remaining fields show the hourly estimates of each
pollutant in grams. The fields are listed as CO 1-24, HC1-24, and NOx 1-24.
3. Scf-em.dbf - The SCF-EM.DBF file is Dbase 4 file of the speed correction factor
running exhaust emission estimates. The file contains the ARID identifier linking
the estimate to the MR and ALLROADS line database. The remaining fields show
the hourly estimates of each pollutant in grams. The fields are listed as CO 1-24,
HC 1-24, and NOx 1-24.
Gridded, Hourly, Emissions Module:
1. Grid-em.dbf - The GRID-EM.DBF file is a Dbase 4 file of the final gridded,
hourly emissions. The file contains an identifier GDID that links to the GRID
174
-------
polygon database. The remaining fields show the hourly estimates of each
pollutant in grams. The fields are listed as CO 1-24, HC1-24, and NOxl-24.
2. Raster Files - The raster datasets created in this module are used for display
purposes and only re-represent data previously estimated. Each file represents a
pollutant-mode-hour specific estimate. Therefore the following raster files are
created:
SZCO1-24
SZHC1-24
SZNO1-24
MRC01-24
MRHC1-24
MRNO1-24
MZCO1-24
MZHC1-24
MZNO1-24
SCFCO1-24
SCFHC1-24
SCFNO1-24
TOT ALCO1-24
TOT ALHC1-24
TOT ALNO1-24
175
-------
TECHNICAL REPORT DATA .
(Please read Isaructions on the reverse before completing}
1. REPORT NO.
EPA-600/R-98-097
2.
3. RECIPIENT'S ACCESSION NO.
4. TITLE AND SUBTITLE
A GIS-Based Modal Model of Automobile Exhaust
Emissions
5. REPORT DATE
August 1998
6. PERFORMING ORGANIZATION CODE
1. AUTHOR(S)
William Hendricks Bachman
8. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Georgia Institute of Technology
School of Civil and Environmental Engineering
Atlanta, Georgia 30332
10. PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
CR823020
12. SPONSORING AGENCY NAME AND ADDRESS
EPA, Office of Research and Development
Air Pollution Prevention and Control Division
Research Triangle Park, NC 27711
13. TYPE OF REPORT AND PERIOD COVERED
Final report; 1/97 - 5/98
14. SPONSORING AGENCY CODE
EPA/600/13
is.SUPPLEMENTARY NOTES APPCD project officer is Carl T. Ripberger, Mail Drop 61, 919/
541-2924.
i6. ABSTRACT
repOrt presents progress toward the development of a computer tool
called MEASURE, the Mobile Emission Assessment System for Urban and Regional
Evaluation. The tool works toward a goal of providing researchers and planners
with a way to assess new mobile emission mitigation strategies. The model is based
on a geographic information syatem (GIS) and uses modal emission rates, varying
emissions according to vehicle technologies and modal operation (acceleration, de-
celeration, cruise, and idle). Estimates of spatially resolved fleet composition and
activity are combined with situation- specific emission rates to predict engine start
and running exhaust emissions. The estimates are provided at user- defined spatial
scales. A demonstration of model operation is provided using a 100 sq km study area
in Atlanta, Georgia. Future mobile emissions modeling research needs are devel-
oped from an analysis of the sources of model error.
17.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
D.IDENTIFIERS/OPEN ENDED TERMS
c. COS AT I Field/Group
Pollution
Exhaust Emissions
Motor Vehicles
Mathematical Models
Information Systems
Geography
Pollution Control
Mobile Sources
Geographic Information
System (GIS)
13 B
2 IB
13 F
12 A
09B.05B
08F
18. DISTRIBUTION STATEMEN1
Release to Public
19. SECURITY CLASS (This Report)
Unclassified
21. NO. OF PAGES
187
2O. SECURITY CLASS (This page)
Unclassified
22. PRICE
EPA Form 2220-1 (9-73)
176
------- |