EP A/600/A-96/130
Integrating Travel Demand Forecasting Models with GIS to
Estimate Hot Stabilized Mobile Source Emissions
William Bachman
Wayne Sarasua
Simon Washington
Randall Guensler
Shauna Hallmark
Michael D, Meyer
NOTICE: The U.S. Environmental Protection Agency through its Office of Research and
Development partially funded and collaborated in the research described here under assistance
agreement CR 817732 to the Georgia Institute of Technology. It has been subjected to Agency review
and approved for publication.
Abstract
In a cooperative research effort with the U.S. Environmental Protection Agency, Georgia Tech
is developing a regional mobile source emissions model using a Geographic Information System (GIS)
framework. The emissions model is designed to improve emission estimates by accounting for the
spatial and temporal effects of a variety of vehicle activities, environmental factors, and vehicle and
driver characteristics. While a description of the overall modeling approach is given, the emphasis of
the paper is to describe the hot stabilized emissions estimation process and the role of travel demand
forecasting models. Although travel demand forecasting models were designed for predicting future
capacity requirements, they also provide useful information needed for mobile source emissions
estimates. Improvements to travel demand forecasting models to more accurately predict hot stabilized
emissions are also discussed.
Introduction
For the past three decades, metropolitan areas in the United States have been using a traditional
four-step modeling process to estimate future demand for transportation. Although not originally
developed for use as part of an air quality modeling package, the four-step process has also become the
standard method for providing a key input, vehicle miles traveled (VMT), into the development of a
mobile source emissions inventory. However, recent research in mobile emissions modeling suggests
that the basic phenomenon being modeled (i.e., emissions from motor vehicles) is much more complex
than can reasonably be represented by the aggregate, network-based four-step modeling approach. In
particular, transportation activities and resulting emissions vary by location and time of day. These
emissions also vary by mode of engine operation (e.g., starts, hot stabilized, enrichment, hot soak
evaporation) that clearly have a spatial relationship to characteristics of the transportation network.

-------
For example, high levels of enrichment emissions are likely associated with long, steep grades where
additional engine power is needed to overcome higher vehicular load.
In a cooperative effort with the United States Environmental Protection Agency (EPA),
Georgia Tech is developing a next generation, mobile source, emissions, research model using a
Geographic Information System (GIS) framework. This approach is designed to improve emission
estimates by accounting for both the spatial and temporal effects of a variety of vehicle activities,
environmental factors, and vehicle and driver characteristics (Bachman et al., 1996). The purpose of
this paper is to discuss the hot stabilized component of the emissions model. The following two
sections present an overview of the research model and provide a rationale for using GIS as the model
platform. The hot stabilized component is presented in detail along with considerations for satisfying
the spatial and attribute data requirements necessary to produce a hot stabilized emissions estimate.
Brief Overview of GIS
A geographic information system is a spatial analysis tool that can be used to model spatial
relationships between geographic entities. A GIS consists of a data base containing spatially
referenced, land-related data as well as procedures for systematically collecting, updating, processing,
and distributing these data. The fundamental structure of a GIS is a uniform referencing scheme which
enables data within a system to be readily linked with other related data.
A true GIS can be distinguished from other graphical database management systems (e.g.,
computer aided design, automated mapping systems) through its capacity to conduct spatial searches
and overlays that actually generate new information. With its robust set of spatial analysis tools, a GIS
can be used to count the number of households that fall within a traffic analysis zone (a point-in-
polygon operation), portion a traffic analysis zone's attribute information to 1 kilometer grid cells
(polygon-overlay-operation), or calculate total road vehicle mileage traveled within a grid cell (line-
through-polygon operation). A vector GIS relies on the topology (the explicit definition of spatial
relationships among entities) of its data structures to do spatial analysis efficiently. Thus, the GIS can
track the relationships among roadway links and readily identify which roadway links share
connections.
Overview of the Model
The model incorporates activity and emission rates associated with specific vehicle operating
modes, such as engine starts, idling, hot stabilized operations, enriched conditions (influenced by high
acceleration and power demand), and hot soak evaporation. The GIS-based model estimates emissions
separately for each operating mode and can aggregate the estimates to a single gridded layer using GIS
spatial analysis tools. The grid cells can then be summed across the different modes to come up with
an overall gridded emission estimate suitable for input into future air quality modeling systems.
The framework for the mobile emissions model is shown in Figure 1. The figure illustrates
how the characteristics of the fleet composition, vehicle activity, and emission rates are used to
2

-------
produce on-network and off-network emissions estimates. Implementing this framework within a GIS
is expected to provide the following benefits:
•	Efficiently manage spatially referenced parameters that affect emissions;
•	Provide manipulation tools to calculate emissions from the modal parameters;
Aggregate emission estimates by mode into separate grid cell layers using topologic overlay
capabilities;
•	Combine the individual modal grid cell layers to obtain an overall mobile emissions estimate for
input into regional air quality models;
•	Provide visualization and map-making tools; and
•	Link to other software packages (such as statistical analysis software or travel demand
forecasting software).
Considerations for Estimating Hot Stabilized Emissions
Several factors affect hot stabilized emission rates: vehicle attributes, environmental conditions,
and vehicle operating modes (Guensler, 1993). Vehicle attributes include 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). Environmental conditions
include ambient temperature, altitude, and humidity. Vehicle operating modes include cruise,
acceleration, deceleration, idle, and induced vehicle loads (e.g., number of passengers, trailer towing,
grade, and air conditioning).
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 MOBILE5a model (USEPA, 1993) and the
California Air Resources Board EMFAC7F models (CARB, 1992), the effects of acceleration,
deceleration, cruise and idle are currently represented by a single surrogate factor: average operating
speed, which is correlated with different proportions of the vehicle operating modes: Vehicle attributes
are 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.
In the mobile emissions model under development, new and improved surrogate variables are
employed (beyond those used in current models) to estimate hot stabilized emission rates. The model
will include additional vehicle operating modes (e.g., the proportion of activity in certain acceleration
regions), additional vehicle attributes (e.g., the percent in acceleration > 3.0 mph/sec), and similar
environmental conditions (e.g., the load distributions and grade). Only those factors that can be
obtained and are correlated with the actual causal variables are selected for inclusion in the hot
stabilized emission rate algorithms. Thus, the model strives to find a balance between optimizing the
accuracy of modeled estimates and using only as many variables as necessary to achieve a functional
model.
3

-------
The hot stabilized portion of mobile emissions is estimated at two levels: a road network level
(major roads) and a mini-zone level (minor roads). The major road network is developed to coincide
with a metropolitan area's travel demand forecasting network. The minor road zones are areas
bounded by major roads. Hot stabilized, emission estimates from both layers are allocated to grid cells
and summed to produce an overall hot stabilized, gridded, emission estimate. Figure 2"shows an
example of an overall hot stabilized emission estimate for Atlanta, Georgia. Darker areas represent
grid cells which have higher portions of hot stabilized emissions.
For each road segment in the major road level, information is stored regarding traffic volume,
subfleet composition, speed/acceleration profile, and other data important for hot stabilized emission
estimation. Road segments modeled at the major road level are not determined by standard road
classifications but by the metropolitan area's travel demand forecasting model's network. This road
segment definition allows travel data produced by the existing planning models to be used in emission
estimation. Use of the travel demand forecasting network indicates that agencies with more
extensively modeled networks could have more accurate spatial emission estimates using this
technique, given that other sources of inaccuracies are equal. Road segment data are used to
determine the quantities and number of seconds of operation of vehicles operating in hot stabilized
mode. The portion of vehicle activity (seconds) for the fraction of fleet operating in hot stabilized
mode is linked to hot stabilized emission rates (grams/second) for each pollutant. Baseline emission
rates will be developed from the EPA MOBDLESb model. To improve model accuracy, activity and
emission rates are tracked for various technology-related fractions of the subfleet rather than for the
subfleet as a whole.
Local roads are modeled on a zonal basis. Each transportation analysis zone (TAZ) is
disaggregated by land use to provide improved spatial resolution of trip origins and destinations. The
trip generation algorithms already employed by the forecasting model can be used to estimate the
number of trips produced by residential neighborhoods (and/or other land use categories) that lie within
a TAZ. The number of trips is multiplied by the estimated average travel time to the closest major road
segment to estimate the vehicle operation within a zone. The output of the local road component of
the hot stabilized module is reconciled with the output from the engine start module to estimate the
portions of vehicle activities before the trip reaches the major road network. The portion of travel time
allocated to each activity will depend on travel times from disaggregated zone cent raids to the major
roads via the local road network.
The individual estimates are aggregated to grid cells using the GIS's spatial analysis
capabilities. Figure 3 shows the spatial allocation technique used to transfer road segment emission
estimates to corresponding grid cells. Grams of hot stabilized emissions are estimated and stored as a
rate per road segment length or zonal area. The spatial entities are subdivided by the grid cell
boundaries and the rates applied to the new spatial structure. All the estimates lying within a grid cell's
boundaries are summed together to produce a grid cell emission estimate.
To estimate light-duty vehicle emissions, the model first estimates the amount of activity
(seconds of operation) on each link and then estimates the fraction of that activity occurring in hot
stabilized enrichment modes. Traffic volumes and speed acceleration profiles are used to estimate the
4

-------
number of seconds of vehicle activity for links in the system. As research continues at Georgia Tech, it
has become evident that the fraction of hot stabilized activity on any road segment is a function of
sub fleet composition and engine load (which is in turn a function of vehicle mass, speed/acceleration
profile, and other load-inducing factors such as grade and accessory operation).
Traffic Volume: Hourly traffic volumes by roadway link will be developed from a calibrated
regional four-step travel demand model. Additional data sources, such as historical Highway
Performance Monitoring System records and real-time data from traffic monitoring systems, will be
used to adjust volumes into hourly volumes throughout a given day. The forecasting model outputs
also provide initial estimates of level of service conditions on the roadway links that are used in
predicting representative speed/acceleration profiles for roadways on the network. The Georgia
Department of Transportation's advanced traffic management system (ATMS) will also provide real-
time traffic volumes and speed acceleration profiles for the monitored freeway and arterial system so
that predicted and monitored conditions can be compared.
Local Fleet Distribution; The composition of the vehicle fleet on any given road segment is
important for two reasons: 1) a small fraction of high-emitting vehicles (super-emitters) must be
correctly represented in the fleet for emissions estimates to be accurate, and 2) the on-road
vehicle/engine combinations determine which fraction of the fleet will be operating under enrichment
conditions for any given speed/acceleration profile, grade, and other influencing factors. The model
aims to group vehicles that are likely to be behave similarly (in terms of basic emission characteristics
and emission response characteristics). Model year and engine size are examples of factors that are
important for characterizing emissions of a vehicle. Clear spatial patterns of vehicle ownership exist in
Atlanta which indicates that on-road subfleet distributions vary as well. Given that on-road fleet
distributions are correlated to the local, sub-regional, and regional registration mix, it is important to
predict various on-road vehicle subfleets for improved spatial resolution. The relationships between
on-road vehicle fleet composition and registered local, sub-regional, and regional fleet composition are
currently being developed through the use of license tag monitoring studies (and translation to vehicle
identification numbers from the department of motor vehicles). Surrounding land use information will
also be used to aid in developing empirical relationships.
Traffic Modal Activity. An engine load module of the modal model employs local subfleet
composition and speed/acceleration profiles (along with grade, and accessory operation assumptions)
to estimate the fraction of vehicles per link operating in enrichment and in hot stabilized mode. The
speed/acceleration operating profile for each roadway is based upon statistically derived relationships
between measured speed/acceleration profiles, level of service conditions for each roadway class, and
physical roadway characteristics. Empirical studies are currently underway to develop modal profile
relationships for freeways and expressways. Work begins this summer on similar arterial and local road
studies. Roadway grades are coded in the model as link attributes and are measured in the field using a
global positioning system array (Awuah-BafFour, et al,, 1996).
5

-------
Travel Demand Forecasting Model Integration
A major goal in the development of the GIS-based mobile emissions model is that the
resulting model be usable in current planning practice. To the extent possible, the model will make use
of existing data sources. Because of this, a linkage to an existing travel model is proposed as an
integral component of the emissions model. Although both technically and practically criticized
(Suhrbier, 1992, Stopher, 1993), travel demand forecasting models are the most widely used and
accepted tools available for providing reasonable forecasts of vehicle activity on a given road network.
Most major metropolitan areas are already using travel demand model outputs in their regional air
quality analyses. Furthermore, most transportation planning agencies within metropolitan areas have
already spent considerable resources developing functional travel demand forecasts using "off-the-
shelf travel demand modeling software. Capitalizing on these initial investments will save
organizations from the painstaking effort of recreating information already available. However, there
are two challenges to integrating a conventional travel demand model into the GIS-based emissions
modeling approach: spatial inaccuracies of the road network and inaccuracies in individual link volume
estimates.
The spatial inaccuracies are the result of the travel demand forecasting model's abstract
representation of an actual road system. A network containing nodes (representing intersections and
shape points) and straight links (representing roadway links) is typically used in travel demand model
highway networks. However, some spatial inaccuracies are observed in these networks, depending on
the level of model detail and whether or not the network was created following an accurate map base.
Because roadway spatial accuracy was not needed for estimating travel behavior, spatial accuracy was
not developed in most urban models. The individual links, however, do maintain a measured road
distance which proves valuable for estimating link volumes.
An analysis of a modeled network for the Atlanta region was completed comparing the traffic
network with estimated traffic volumes of a TRANPLAN model and a spatially accurate digital road
network (Figure 4). A sample area of 182 square kilometers was used in the comparison. Conflation
(matching links between networks so that attributes can be transferred) techniques were used to
allocate the TRANPLAN forecasted volumes to the accurate digital road network. The simplification
of using traffic volumes instead of emissions does not affect the findings because hot stabilized link
emissions are roughly proportional to the traffic volume on the link. The volumes associated with the
TRANPLAN network, and the corresponding volumes associated with a spatially accurate network
were aggregated to 1 km grid cells in the GIS using a line-through-polygon procedure. Results show
that the TRANPLAN volume estimates by grid cell on average were within 30 percent of the
corresponding grid cell estimates using the accurate digital map. Individual cell estimates, however,
varied widely. The TRANPLAN link, for example, could be far enough off spatially such that most of
its volume could be attributed to the incorrect cell. This error can seriously affect the confidence in a
single cell's estimate. The aggregation steps were repeated for a 4 km grid cell aggregation in order
measure the sensitivity to grid cell size. As one would expect, a larger grid cell is able to reduce the
spatial error. An aggregation over the entire study area showed total volumes to be within 2 percent of
the more accurate road database.
6

-------
As a result, using existing travel demand forecasting model networks in emission modeling
reduces the accuracy which can be expected for individual cells. However, inaccurate cell estimates are
frequently balanced by an adjacent cell's error. A better test of the sensitivity and level of acceptable
inaccuracy would be to develop hot stabilized emission estimates for both the abstract netwdrk and the
accurate network and input the results into a photochemical model. The emission model and the EPA's
future MODELS3 photochemical model are still under development, and this experiment will be run
when the models are complete. In the meantime, a test is underway using the sample gridded estimates
in an urban airshed model to determine the impacts that counterbalancing spatial errors have on
predicted ozone concentrations.
If improvements to the spatial accuracy of the network representation in the travel demand
forecasting model are deemed necessaiy (based upon air quality model sensitivity analysis), two
alternatives could be implemented. The first is to improve the accuracy of the highway network by
adding shape points and/or more accurately representing complicated roadway configurations such as
those at freeway interchanges. This network-editing process would be quite time consuming
depending on the size and accuracy of the original model network. The alternative is to conflate the
links of the travel demand modeling network to an accurate digital road network. Conflation involves
the development of a one-to-one linkage between a model link and a corresponding road segment on
the accurate digital road network. Emissions could be estimated for the existing travel demand model
network and then transferred by unique identification number to the corresponding segments of the
road database for aggregation to grid cells. The disadvantage to this strategy is that conflation can also
be a labor intensive process for large urban networks. Conflation tools and techniques are available to
make the process less intensive.
The inaccuracy of the travel demand model's link volume estimates is another issue that must
be considered. Spending a great deal of time and effort to improve a model's spatial accuracy is not
justifiable unless the modeled volumes are accurate, which is usually not the case on a link-by-link
basis. Travel demand models are best suited for predicting corridor flows among large areas. While
total volumes on parallel links in one general direction are usually within 10 percent of actual volumes
during a calibration base year, volumes on a link-by-link basis are not nearly as accurate. If base year
link volumes predicted by a model are not accurate, it is even more unlikely that forecasted volumes for
individual links are accurate.
Conclusion
A next generation research mobile emissions research model is being developed which will
increase the spatial resolution of mobile source emission inventory predictions. The hot stabilized
portion of the emissions modeling framework relies on the integration of travel demand and emissions
forecasting models to provide information regarding travel behavior. Travel demand models contain
the spatial variations in travel behavior that allow hot stabilized emissions to be estimated with greater
spatial accuracy. Integrating existing travel demand models is advantageous as information already
produced at great expense to planning agencies can be used.
7

-------
References
Awuah-Baffour, Robert, Wayne Sarasua, Karen Dixon, and William Bachman; GPS with an Attitude;
Proceedings of the 1996 AASHTO GIS-T Symposium (forthcoming); Kansas City, MO; March 1996
Bachman, William, Wayne Sarasua, and Randall Guensler; GIS Framework for Mobile Source
Emissions Modeling; Transportation Research Record: Number (forthcoming); pp. (forthcoming);
Transportation Research Board; Washington, DC; 1996
CARB; California Air Resource Board; Methodology to Calculate Emissions Factors for On-Road
Motor Vehicles; Technical Support Division; Sacramento, CA; 1992
Guensler, Randall; Vehicle Emission Rates and Average Vehicle Operating Speeds; Dissertation;
submitted in partial satisfaction of 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
Stopher, P.R.; Deficiencies of Travel-Forecasting Methods Relative to Mobile Emissions; Journal
of Transportation Engineering. ASCE; Vol, 119, No. 5; September/October 1993
Suhrbier, John H.; Opportunities for Improved Transportation Planning; Transportation Planning
and Air Quality; pp 30-45; ASCE; 1992
USEPA; User's Guide to MOBILES; U.S. Environmental Protection Agency, Office of Mobile
Sources; March 1993
8

-------
Emissions
Vehicle
Activity
Fleet
Composition
Emission
Rates
Super
Emitters
Normal
Emitters
Distributions:
Drivers
Vehicle Class
Model Year
Accrued VMT**
Fuel Delivery System
Engine Displacement
Gross Vehicle Weight
Emission Controls
Computer Controls
Tampering
1AM Hjstofy
Towing
Air Conditioning
Traffic
Volumes
Modal
Profiles
Speed
Acceleration
Modal
Emission
Rates
Driver
Behavior
Road
Grade
Facility
Design
Nomul-ErniQef and Super-Emitter
Modal Emission Rjte Algorithms for
Hoi-Stabilized tnd Enrichment Activities
Are Distinct and Employ • Vtriety of
Vtriabks tnd Interactions
Engine Start
Profiles
Soak
Time
Modal
Profile
Network
Access Time
Evaporative
Emission
Rates
Modal Start
Emission
Rates
Link Emissions
(On-Network)
VHT* Hot-Stabilized
VHT Enrichment
Engine Starts Component
VHT Running Losses
Zone Emissions
(Off-Network)
Engine Starts
VHT Hot-Stabilized
VHT Enrichment
Hot Soaks
Diurnal Evaporative
VHT Running Losses
(*)VHT = Vehicle hours traveled
(**)VMT = Vehicle miles traveled
FIGURE 1 - GIS-based emission model framework.
9

-------
Emissions by Grid Cell
Hi	Highest 1%
Hi	Highest 1-10%
IB	Highest 10-25%
H	Highest 25-50%
II	Lowest 25-50%
HI	Lowest 25%
V..&W
St

v V
A5!1!# «
ymg--- »
m
^ & Me
K $
¦c- yM*
& S? *

w I *»

«N ¦»
p§

¦ :•:«« W
, : JS S*


rniimt?
«... ^ ^

ft:$S £ -s
& ^	-3S.wi
SS :• £
~4-
Grid Cell = 1 km x 1
w/jK>iS
FIGURE 2 - Estimated carbon monoxide produced by hot-stabilized emission
activity during a mid-week AM peak hour in the Atlanta, GA area.
10

-------
Convert grams
of CO to
grams of CO
per mile
All segments
have same
grams of CO
per mile
Convert grams of CO
per mile back into grams
by multiplying by length
of segment
Sum all line
segments within
the boundaries of
each grid cell
FIGURE 3 - Spatial aggregation technique for moving linear emission data to grid cells.
11

-------
0
FIGURE 4 - Sample of the TRANPLAN highway network (gray) for Atlanta and
the corresponding spatially accurate highway network (black).
12

-------
MD, n iT-n -n nc TECHNICAL REPORT OATA
IN K1V1K 1-j-K 1 r r ilD (Please read Instructions on the reverse before completing)
—
1, REPORT NO. 2.
EPA/600/A-96/130
3. REC
4. TITLE AND SUBTITLE
Integrating Travel Demand Forecasting Models with
GIS to Estimate Hot Stabilized Mobile Source
Emissions
5. REPORT DATE
6. PERFORMING ORGANIZATION CODE
?.author(s)^i Bachman, W. Sarasua, S.Washington,
R. Guensler, S. Hallmark, and M. D. Meyer
8. PERFORMING ORGANIZATION REPORT NO.
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Georgia Institute of Technology
School of Civil and Environmental Engineering
790 Atlantic Avenue
Atlanta, Georgia 30332-0355
10. PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
CR817732
12. SPONSORING AGENCY NAME ANO 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
Published paper; 1/95—3/96
14. SPONSORING AGENCY COOE
EPA/600/13
15. supplementary NOTES APPCD pro:ject officer is Carl T> Ripberger. Mail Drop 61, 919/
541-2924. Presented at AASHTO GIS-T Conference, Kansas City, MO, 3/31/96.
i6. abstract paper discusses integrating travel demand forecasting models with the
Geographic Information System (GIS) to estimate hot stabilized mobile source emis-
sions. In developing a regional mobile source emissions model using a GIS frame-
work, the emissions model is designed to improve emission estimates by accounting
for the spatial and temporal effects of a variety of vehicle activities, environmental
factors, and vehicle and driver characteristics. While a description of the overall
modeling approach is given, the emphasis of the paper is to describe the hot stabili-
zed emissions estimation process and the role of travel demand forecasting models.
Although travel demand forecasting models were designed for predicting future capa-
city requirements, they also provide useful information needed for mobile source
emissions estimates. Improvements to travel demand forecasting models to more
accurately predict hot stabilized emissions are also discussed.
17. KEY WORDS AND DOCUMENT ANALYSIS
a. DESCRIPTORS
b.IDENTIFIERS/OPEN ENDED TERMS
c. COS ATI Field/Group
Pollution Travel
Ground Vehicles Forecasting
Emission Carbon Monoxide
Automobiles
Vehicles
Methematical Models
Pollution Prevention
Mobile Sources
Automotive Vehicles
13 B 05F
13 F
07B
14G
12 A
18. DISTRIBUTION STATEMENT
Release to Public
19. SECURITY CLASS (This Report)
Unclassified
21. NO. OF PAGES
20. SECURITY CLASS (This page)
Unclassified
22. PRICE
EPA Form 2220-1 (9-73)

-------