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

-------