Quick Response Freight Manual II

4.0 Incorporating Freight into “Four-Step” Travel Forecasting

This section explains the various methods of incorporating freight into the traditional “four-step” travel forecasting process. The four steps include trip generation, trip distribution, mode choice, and trip assignment. These are explained in more detail in the ensuing sections. The focus will be at three levels of geography – urban, statewide, and site specific.

4.1 Introduction

The flow of freight can be measured in two forms – commodity and trucks. Figure 4.1 depicts the four steps to forecasting freight at any geographic level. As indicated in the initial steps, trip generation and distribution can either be in the form of commodities or trucks. The basic difference between commodity- and truck-based models is the form of the input data. However, for trip assignment purposes all forms of freight are converted to vehicles to be assigned onto a roadway network.

Figure 4.1 "Four-Step" Process of Freight Forecasting

The following subsections discuss the general issues of incorporating freight into tradition four-step transportation models, a topic that is discussed in detail in later sections as it applies to urban and state models.

4.1.1 Trip Generation

Trip generation uses economic variables to forecast freight flows/vehicle flows to and from a geographic area using equations. The trip generation equations are either borrowed from other sources or developed locally by using an existing commodity flow table or by estimating from vehicle surveys. The outcome of trip generation is the amount of a commodity and/or the number of vehicles that comes into or goes from a particular geographic unit in a specified unit of time.

Trip generation models used in freight forecasting include a set of annual or daily trip generation rates or equations by commodity. These rates or equations are used to determine the annual or daily commodity flows originating or terminating in geographic zones as a function of zonal or county population and/or industry sector employment data. In other words, employment and/or population data are the essential input data required for computing freight trip generation.

The independent variables, such as employment and population, usually dictate the level of detail the freight flows can be generated using a trip generation model. This may be a county or a traffic analysis zone (TAZ). The travel demand models usually use TAZ data, and so a freight forecasting model can be developed at a TAZ level as long as the base and forecast year data at the required level of industry detail is available at that geographic unit.

Before trip generation models are estimated, trucks are first classified by type of truck and/or trip purpose/sector. The various types of classification of trucks include the FHWA classification system, gross vehicle weight (GVW) ratings, type of goods carried, number of tires/axles, and body type.

Normally, one set of regression equations for the productions and one set of regression equations for consumption are estimated. These regression equations are either developed for each commodity group or truck type. A commodity group is analogous to a “trip purpose” in passenger modeling. The intercept is almost always forced to zero, because there should be no freight in or out of a zone with no related economic activity. The observations used to estimate the regression model would be the inbound tons of the commodity or number of trucks and the independent variables are usually employment, industry type, population, etc. for each geographic area.

Truck trip generation rates can be developed from trip diary surveys using regression equations by regressing the number of commercial vehicles on the number of employees in various industries and household population. Trip rates also can be estimated for each individual land-use type based on the ratio between the truck trips coming into and going out of the land area and the employment associated with that land use. The 1996 Quick Response Freight Manual (QRFM) was developed by the FHWA and it provides default values that can be used in models. The QRFM rates were developed using regression models developed from a trip diary in Phoenix. The NCHRP Truck Trip Generation Synthesis (298) is another source for a complete reference list of potential trip rates.

The various steps required to determine trip rates are:

Trip rates need to be estimated or identified (either through local surveys or using national default data);
Socioeconomic data (employment by industry and households/population) by TAZ is applied to the rates to get generation by TAZ;
The QRFM method assumes that productions equal attractions, but local data can be used to estimate separate production and attraction rates; and
If there are freight centers (ports, intermodal terminals), they should be treated as special generators and have their own trip rates determined from surveys since employment rates would not apply.

Table 4.1 is borrowed from the Phoenix Metropolitan Urban Truck Model. [Earl Ruiter; Cambridge Systematics, Inc.; Development of an Urban Truck Travel Model for the Phoenix Metropolitan Area; February 1992; Report Number FHWA-AZ92-314; prepared for Arizona Department of Transportation and the Federal Highway Administration.] There are far more four-tire truck trips per unit of activity than combination and large-truck trips, which is pretty typical in an urban area. It should be noted that households also do generate a lot of truck trips.

Table 4.1 Truck Trips Rates

Generation Variable (Employment or Households)	Four-Tire Trucks	Single Unit Trucks (6+ Tires)	Combination Trucks
Agriculture, Mining, and Construction	1.110	0.289	0.174
Manufacturing, Transportation/ Communications/Utilities, and Wholesale	0.938	0.242	0.104
Retail Trade	0.888	0.253	0.065
Office and Services	0.437	0.068	0.009
Households	0.251	0.099	0.038

4.1.2 Trip Distribution

In trip distribution, one determines the flow linkages between origin and destination for those commodity tons/truck trips that were developed in trip generation. Trip distribution uses those flows/trips to and from and independent variables on the transportation system to forecast the flows/trip interchanges between geography areas.

The trip distribution equations can be borrowed from other sources or developed locally by using an existing commodity flow table or local vehicle surveys. A gravity model can be constructed and calibrated at a prespecified geographic detail. The gravity model is a statistical process that has been found useful to explain the relationship between transportation zones. The considerations are the total trips that begin in the first zone, the number ending in the second zone, and the impedance or difficulty to travel (such as cost or time) between them.

The average trip lengths needed to obtain trip-length frequency distributions and friction factors are normally obtained from surveys. The degree of difficulty of travel, usually a function of some impedance variable used in the distribution model needs to match the survey data (free flow time, congested travel time) and there must be a source of the impedance variable. The calculation of the degree of difficulty is often called a friction factor. With limited survey data, the model is typically calibrated at the district level, and the friction factors developed are assumed to apply at smaller units of geography. However, it is sometimes difficult to get survey data for trip distribution, and friction factors are often borrowed from other sources.

The friction-factors are usually calculated as a negative exponential function of the average trip time from origin TAZ to destination TAZ. The parameters in the exponential function are calculated from the trip length frequency distribution, which describes the shape of the curve that is summarized by the average trip length.

The friction factor curves for the PSRC truck model [Cambridge Systematics, PSRC Model Improvements, 2002] were derived originally from the 1996 edition of the (QRFM [Cambridge Systematics, Inc., Quick Response Freight Manual, Federal Highway Administration, 1996] and adjusted to provide the best fit with the average trip lengths from the origin-destination survey of trucks. The light, medium, and heavy trucks are distributed from origins to destinations using this gravity model technique with different parameters. These friction factors were developed using impedance functions that also varied by trip distances, that is different parameters were used for short and long distances, as shown below:

Light impedance function:
- exp (3.75 – 0.08 * light truck generalized cost skim) for less than 26 miles
- exp (2.1 – 0.005 * light truck generalized cost skim) for greater than or equal to 26 miles
Medium impedance function:
- exp (4.75 – 0.05 * medium truck generalized cost skim) for less than 27 miles
- exp (4.2 – 0.003 * medium truck generalized cost skim) for greater than or equal to 27 miles
Heavy impedance function:
- 1.0 for less than 7.5 miles
- exp (5.0 – 0.009 * heavy truck generalized cost skim) for greater than or equal to 7.5 miles

Table 4.2 shows the average trip lengths from the PSRC truck model compared against the observed trip lengths.

Table 4.2 Average Truck Trip Lengths

Trip Length	Light Truck	Medium Truck	Heavy Truck
Observed Trip Length (Miles)	No data	27.51	30.81
Modeled Trip Length (Miles)	22.34	27.53	28.29

Another method that is less popular is the growth factor approach for trip distribution, also known as the Fratar method. This usually requires an existing base year trip table of freight flows or trip interchanges. The Fratar method assumes that the change in the number of trips in an O-D pair is directly proportional to the change in the number of trips in the origin and destination. The method lacks system sensitivity to the change in network-level characteristics such as congestion. Also, these methods allow preservation of observations as much as is consistent with information available on growth rates. If part of the base year matrix is unobserved, then this error is carried over in the forecasts. These methods cannot be used to fill in unobserved cells of partially observed trip matrices. Hence, they are of limited use to test new policy options.

4.1.3 Mode Split/Conversion to Vehicle Flows

Mode choice modeling is used if multimodal trip tables need to be prepared. This step allows the forecastability of mode splits as they change over time. The four major categories in which various factors that affect mode choice decision-making process fall into are:

Goods Characteristics – These include physical characteristics of goods such as the type of commodity, the size of the shipments, and the value of the goods;
Modal Characteristics – Speed of the mode, mode reliability, and the capacity;
Total Logistics Cost – Inventory costs, loss and damage costs, and service reliability costs; and
Overall Logistics Characteristics – Length of haul and the shipment frequency.

Figure 4.2 shows the major characteristics of each of the freight modes in a continuum/spectrum and shows how this relates to the types of goods that may be shipped by each mode. The rail and water modes have the highest capacity on this spectrum, while air and truck have the lowest capacity. The air and truck modes provide the highest level of service in terms of reliability and minimal loss and damage. So commodities that are needed for just-in-time production systems (like certain machinery parts) will need to use trucking and air. The mode associated with the highest cost is by air and, therefore, are only justified for high-value commodities such as electronics.

Figure 4.2 Goods and Modal Characteristics

The two common methods of computing mode splits are the market segmentation method and the choice method. The market segmentation method is described in detail later in this section.

Choice Method

These methods are the most comprehensive as they examine the characteristics of each individual shipment and the available modes. The most common type of choice method is the discrete choice logit model. This formulation is very similar to the passenger mode choice modeling, but the variables and data sets used to estimate the parameters are very different. The logit discrete choice model shows the choices for individual shipments as a function of the utility that each mode provides to the shipper. Utility can be a function of any of the factors mentioned earlier in this section.

The logit model actually calculates the probability that each shipment will use a particular mode. Summing the probabilities across all of the shipments provides the overall mode share. Each modal alternative has a utility to the shipper that has a systematic component related to the factors we have described earlier and a random component that has to do with things like personal relationships. The coefficients in the utility function measure the relative importance of each factor in determining mode choice. The greater the utility that any alternative has, the higher the probability that this alternative will be selected.

Logit choice models are the most complete with respect to modeling all of the factors that affect mode choice. Thus, they can be applied to a wide range of policy and investment studies. However, they are complex to build and are very data intensive. Most of the data needed require the use of complex performance or simulation models. The truck surveys are helpful for estimating the choice parameters, but these surveys are expensive and time-consuming to conduct.

Truck Conversion

The freight trip tables after the mode split step are multimodal commodity flow tables in annual tons. That is, after allocating the tables among the modes, the flow units will still be in annual tons. The flow unit in almost all highway travel demand models is daily or peak-period vehicles. Therefore, to consider the interaction of freight trucks on the highway with all automobiles and all other vehicles, the time period must be made consistent and the annual truck tables in tons must be converted from annual tons to daily trucks. Payload factors (average weight of cargo carried) are used to convert tons to trucks. The annual trips are then converted to daily trips by assuming an average number of operating days per year. But most travel demand models use average weekday travel. Various data sources can be used to estimate fraction of truck tonnage on weekdays and then divide this tonnage by number of weekdays per year. This process is discussed in more detail in later in this section.

Payloads or truck loads are limited by weight and volume considerations. The commodities carried by trucks have different densities and, therefore, different payloads for the same volume. Because of handling and packaging needs, payloads also may differ by commodity. For example, large size trucks carry heavier loads even for the same commodity. If payloads are calculated for different truck classes, the commodity tonnage needs to be allocated to the different truck classes. Smaller trucks tend to be used more in shorter-haul service. To the extent that length of haul and truck size are correlated, length of haul (directly available from commodity flow data) can be used in calculating payload factors. Payload factors can be calculated for loaded trucks only (estimated truck volumes must then be adjusted to account for percent of empties) or they can average empty and loaded weights.

The various sources of payload factors are 1) shipper or carrier surveys that provide information about the tonnage and commodity being carried; 2) weigh stations that typically have weight information by truck type, but not by commodity; and 3) the VIUS is a part of the Economic Census and is collected every five years. ["VIUS" stands for Vehicle Inventory and Use Survey, U. S. Census Bureau, 2002. The survey was first conducted in 1963, under the name of Truck Inventory and Usage Survey (TIUS). It was renamed as VIUS in 1997. The survey was discontinued after the 2002 survey year was processed. It had been conducted for years ending in “2” and “7.”]

4.1.4 Network Assignment

The process of allocating truck trip tables or freight-related vehicular flows to a predefined roadway network is known as the traffic assignment or network assignment. There are many types of assignments that are dependent on a number of factors such as level of geography, number of modes of travel, type of study and planning application, data limitations, and computational power such as software. The various types of assignments and their applications are explained in detail later in this section.

In developing a truck trip assignment methodology, some of the key issues and model components that need to be addressed and evaluated are as follows:

Time-of-Day Factors – These distribution factors by truck type separate truck trips that are in motion during each of the four modeling time periods; these factors need to be examined through recent data.
Roadway Capacity and Congested Speeds – A single truck will absorb relatively more of the available capacity of a roadway than an automobile, and a given volume of trucks will often result in a much greater impact on congested speeds than a similar volume of automobiles. So passenger car equivalent (PCE) factors are required to convert the truck flows to PCEs before the assignment process.
Volume-Delay Functions – These functions are used to estimate average speeds as a function of volume and capacity may be different for trucks than for automobiles.
Truck Prohibitions – Some freeways and major principal arterials in the region have prohibitions for certain classes of trucks, and this needs to be addressed before the assignment. A truck network also may be built based on the local knowledge of truck prohibitions and truck routes.

4.2 Urban Freight and Commercial Trucks

4.2.1 Definition of Trucks

In order to capture trucks accurately in a truck-travel model system, the mode “truck” needs to be defined first. This can be accomplished by examining the different types of trucks and identifying the different types of truck classification variables in the region. This essentially involves the way a truck is defined by its physical characteristics. This section describes the various classification variables that have been widely used by various agencies.

Number of Axles

The total number of axles on the trucks are normally categorized into four axle categories – two axles with four tires, two axles with six tires, three axles, and four or more axles. This information on vehicles can be obtained by visual identification or manual counts, or the use of axle sensor-based counters that are often used to collect accurate truck counts. The number and spacing of axles is used to classify trucks into FHWA’s 13-category classification scheme. Most of the vehicle classification count studies across the country classify trucks into these 13 categories, as listed below:

Class 1: Motorcycles (Optional) – All two- or three-wheeled motorized vehicles. Typical vehicles in this category have saddle type seats and are steered by handlebars rather than steering wheels. This category includes motorcycles, motor scooters, mopeds, motor-powered bicycles, and three-wheel motorcycles. This vehicle type may be reported at the option of the state.
Class 2: Passenger Cars – All sedans, coupes, and station wagons manufactured primarily for the purpose of carrying passengers and including those passenger cars pulling recreational or other light trailers.
Class 3: Other Two-Axle, Four-Tire Single Unit Vehicles – All two-axle, four-tire vehicles, excluding passenger cars. Included in this classification are pickups, panels, vans, and other vehicles such as campers, motor homes, ambulances, hearses, carryalls, and minibuses. Other two-axle, four-tire single-unit vehicles pulling recreational or other light trailers are included in this classification. Because automatic vehicle classifiers have difficulty distinguishing Class 3 from Class 2, these two classes may be combined into Class 2.
Class 4: Buses – All vehicles manufactured as traditional passenger-carrying buses with two axles and six tires or three or more axles. This category includes only traditional buses (including school buses) functioning as passenger-carrying vehicles. Modified buses should be considered to be a truck and should be appropriately classified.
Class 5: Two-Axle, Six-Tire, Single-Unit Trucks – All vehicles on a single frame, including trucks, camping and recreational vehicles, motor homes, etc., with two axles and dual rear wheels.
Class 6: Three-Axle Single-Unit Trucks – All vehicles on a single frame, including trucks, camping and recreational vehicles, motor homes, etc., with three axles.
Class 7: Four-or-More-Axle Single-Unit Trucks – All trucks on a single frame with four or more axles.
Class 8: Four-or-Fewer-Axle Single-Trailer Trucks – All vehicles with four or fewer axles consisting of two units, one of which is a tractor or straight truck power unit.
Class 9: Five-Axle Single-Trailer Trucks – All five-axle vehicles consisting of two units, one of which is a tractor or straight truck power unit.
Class 10: Six-or-More-Axle Single-Trailer Trucks – All vehicles with six or more axles consisting of two units, one of which is a tractor or straight truck power unit.
Class 11: Five-or-Fewer-Axle Multitrailer Trucks – All vehicles with five or fewer axles consisting of three or more units, one of which is a tractor or straight truck power unit.
Class 12: Six-Axle Multitrailer Trucks – All six-axle vehicles consisting of three or more units, one of which is a tractor or straight truck power unit.
Class 13: Seven-or-More-Axle Multitrailer Trucks – All vehicles with seven or more axles consisting of three or more units, one of which is a tractor or straight truck power unit.

Gross Vehicle Weight (GVW)

GVW is a unique characteristic of a vehicle that is the sum of the empty vehicle weight and its payload. GVW classification ratings are primarily used for air quality modeling purposes. GVW ratings of vehicles cannot be observed or measured but can only be determined while administering intercept surveys. Hence, it is hard to associate a vehicle of certain GVW to a particular FHWA vehicle configuration as it only gives an indication about probable body type or even vehicle configuration. EPA provides guidance on the mapping of FHWA vehicle classes to MOBILE 6 vehicle classes for air quality modeling. [http://www.epa.gov/ttn/chief/eiip/techreport/volume04/ (see PDF of Chapter 2).] The VIUS database also provides a correlation between number of axles and GVW, and the GVW classes included in VIUS are: less than 6,000 pounds; 6,001 to 10,000 pounds; 10,001 to 14,000 pounds; 14,001 to 33,000 pounds; and greater than 33,001 pounds.

Vehicle Configuration – This is primarily based on the physical appearance of a vehicle. The classification scheme adopted by FHWA separates vehicles into 13 categories depending on whether the vehicle carries passengers or commodities. Nonpassenger vehicles are further subdivided by number of axles and number of units both power and trailer units. The VIUS database also has information on vehicle configuration but it classifies vehicles into four more general categories than the FHWA 13 vehicle classes. It also provides information on the axle arrangement, i.e., truck type and number of axles on a particular truck and/or combination. This variable in VIUS has more detail to the tune of 72 vehicle classes. So the correlation between FHWA and VIUS classifications is not very strong in terms of a perfect match.
Length of Vehicle – The length of a vehicle also is an important variable of interest if it can be measured accurately. The counters recommended by the traffic monitoring guide use two inductance loops to estimate length of vehicles crossing the loops. These dual loop sensors are generally capable only to classify vehicles into fewer categories than the FHWA 13 vehicle classes. The VIUS database reports the overall length of the vehicle or vehicle and trailer as it was most often operated.
Body Type – This type of classification is based on the appearance of the body of the vehicle and the type of commodity it carries most often. The Department of Motor Vehicles (DMV) data classifies vehicles based on body type. The California DMV data from the California Energy Commission that was used for the Southern California Council of Governments (SCAG) Heavy Duty Truck (HDT) Model Update classifies vehicles into about 55 categories and has a correlation with the GVW ratings. The VIUS adopts a different body type classification (32 classes) that is quite different from the DMV database. This type of information can be gathered only by visual or manual observations. Also, the plethora of body types makes it hard to correlate it to any other classification system.

The definition and classification of trucks into appropriate categories are very important so that accurate and reliable data is modeled to produce good forecasts. Hence, a proper classification system that is consistent across all the data sources should be developed. It is not just enough if a proper classification system is identified when developing a truck model, but also should ensure that observed data within the same classification system can be collected to validate the truck model against.

The SCAG HDT model represents heavy-duty trucks only, that is, trucks that are over 8,500 pounds. The primary use of this model is for air quality purposes and so it uses the weight-based classification system. These are:

Light-heavy (8,500 to 14,000 pounds);
Medium-heavy (14,000 to 33,000 pounds); and
Heavy-heavy (greater than 33,000 pounds).

The PSRC truck model also classifies trucks based on weight but these categories also are loosely correlated to other defining characteristics of trucks for other purposes. These are:

Light Trucks – Four or more tires, two axles, and less than 16,000 pounds (this also includes nonpersonal use of cars and vans);
Medium Trucks – Single unit, six or more tires, two to four axles and 16,000 to 52,000 pounds; and
Heavy Trucks – Double or triple unit, combinations, five or more axles, and greater than 52,000 pounds.

The San Joaquin Valley truck model in central California is designed to generate truck volumes based on truck classes that the California Air Resources Board defines as medium-heavy and heavy-heavy duty for regulatory purposes (more than 14,000 pounds gross vehicle weight rating). These are:

Medium-Heavy Duty Trucks – GVW rating between 14,001 and 33,000 pounds; and
Heavy-Heavy Duty Trucks – GVW rating of 33,001 pounds and more.

The current Maricopa Association of Governments (MAG) truck model is based on GVW as well that includes three classes – light (less than 8,000 pounds), medium (8,000 to 28,000 pounds), and heavy (greater than 28,000 pounds). As the vehicle classification counts are based on FHWA classes, and due to the difficulty in correlating the GVW classes to FHWA classes, the new MAG truck model will include three groups of trucks. These are based on the FHWA classification system, as shown below:

Class 3 – 2-axle, 4-tire commercial vehicles (“Light”);
Classes 5-7 – 3+ axle, 6+ tire, single unit commercial vehicles (“Medium”); and
Classes 8-13 – 3+ axle, 6+ tire, combination unit commercial vehicles (“Heavy”).

4.2.2 Trucks that Do Not Carry Freight

There is a unique segment of truck population that does not carry freight, which also is known as the service sector. This includes trucks that are used in the utility sector and other services related to commercial and residential land uses (i.e., business and personal services). Data on this type of trucking activity is difficult to collect through conventional survey methods because of overlapping nature of these types of truck trips with other industry types. As part of the FHWA commercial vehicle study, a method was developed based on various data sources that are commonly available to an agency. This methodology is provided in this section.

Model Methodology

If a separate model is to be created for trucks that do not carry freight, then it may be necessary to conduct a survey of the activity of these types of trucks. Without such a survey, it may be extremely difficult to update or calibrate this part of the truck model. There was data collected as part of the FHWA research on accounting for commercial vehicles in urban transportation models [Cambridge Systematics, Inc., Accounting for Commercial Vehicles in Urban Transportation Models, prepared for Federal Highway Administration, February 2004] that identified the magnitude and distribution of service vehicles in four categories: safety, utility, public service, and business and personal service vehicles. Data from the California DMV was used to identify fleet sizes for these vehicles. Average daily trip lengths were identified for these vehicles from the 2002 VIUS, which was summarized for metropolitan areas. VIUS also can be summarized by state or metropolitan areas within a state, but this may be too small a sample size. A similar approach currently is being proposed in the Phoenix MPO, the MAG, truck study where the size and weight of the vehicles in this category will be determined from the MAG region DMV registration data. In the event of lack of DMV data, truck population data by FHWA classes will be derived from the most recent county-by-county estimates of trucks from MAG’s Air Quality Planning department.

The four types of service vehicles in an urban metropolitan area are:

Safety vehicles;
Utility vehicles;
Public service vehicles; and
Business and personal service vehicles.

Public service vehicles are publicly owned. Business and personal service vehicles are privately owned. Safety and utility vehicles may be either publicly or privately owned.

About 5.9 percent of the total vehicle miles traveled in the urban areas in the United States each year is attributable to vehicles in these four categories. Business and personal-service vehicles alone contribute 3.6 percent of the total VMT in urban areas across the nation, while public-service vehicles contribute 1.6 percent of the total VMT and safety and utility vehicles contribute 0.4 percent each.

Many older urban transportation models currently do not include specifically include commercial service vehicles, although some models have identified a commercial vehicle trip purpose that is based on a fixed factor of personal nonhome-based travel. Some truck models also include delivery and service vehicles that are four-tire commercial vehicles, based on the inclusion of these vehicles in the 1996 edition of the Quick Response Freight Manual.

Data Sources

One of the key sources of information essential for estimating a model for this sector is the truck populations for the four categories of service vehicles. DMV registration data and commercial vehicle surveys have been usefud to estimate truck populations for this sector. These are described below.

Cambridge Systematics, Inc. (CS) created a dataset combining data on safety, utility, public service, and business and personal service vehicles.

Safety vehicles were derived from two sources: 1) California DMV data on police, fire and rescue vehicles, and tow trucks for Los Angeles, San Francisco, San Diego, and Sacramento; and 2) the Detroit commercial vehicle survey, which includes snow plows and tow trucks.
Utility vehicles were derived from two sources: 1) California DMV data on utility cars and trucks, water and irrigation trucks, and garbage trucks for Los Angeles, San Francisco, San Diego, and Sacramento; and 2) three commercial vehicle surveys that included utility and maintenance vehicles for the Detroit, Atlanta, and the Triad cities regions.
Public service vehicles were derived from a single source: California DMV data on city, county, state, Federal, other, and school and college cars for Los Angeles, San Francisco, San Diego, and Sacramento.
Business and personal service vehicles were derived from two sources: 1) California DMV data on “other commercial cars,” armored, panel and pickup trucks, vans and step vans for Los Angeles, San Francisco, San Diego, and Sacramento; and 2) three commercial vehicle surveys that included vehicles used for office, professional, or personal services in the Detroit, Atlanta, and Denver areas.

Data for four cities – Los Angeles, San Francisco, San Diego, and Sacramento – were compiled and analyzed because these were the only four cities with a comprehensive assessment of all commercial service vehicles. Demographic data for each city, including total population and employment by type (government, utility, business and personal services, and total), were derived from the 2000 Census.

For the new MAG truck model update, a new approach on deriving this data is being proposed due to the lack of DMV data. The truck population data and the VMT distributions at the county level is being prepared before estimating parameters for this sector. CS obtained the truck population data at the county level for all the counties in the State of Arizona. These data are at the 13 FHWA classes and will be disaggregated to the 28 MOBILE6 vehicle categories to get a better sense of the body type of trucks. This disaggregation process will be based on the VMT mix data for the 28 vehicle classes that already are derived for air quality modeling purposes at MAG. For the FHWA research project, CS developed a method that correlates body type of trucks to the use of the truck or industry sector. This method will be used here to identify those vehicle classes out of the 28 that fall under the service industry sector.

Aggregate Demand Method

The Aggregate Demand Method estimates service vehicle fleet size based on two demographic factors: total employment (possibly stratified by type) and population. A summary of the travel behavior characteristics is provided in Table 4.3. This summary includes estimates of fleet size, number of trips, and VMT calculated from a statistical analysis of the available data combined with demographic data. The only comprehensive data source (including both public and private sector data) is the motor vehicle registration data, so only these data are used in estimating rates of travel by commercial service vehicles. The data shown in Table 4.3 do not show trips per vehicle, so the commercial vehicle surveys from other cities are used to provide data on this variable for private sector vehicles only. The percent of vehicle miles traveled will be derived from MAG’s air quality modeling work.

Table 4.3 Travel Behavior Characteristics for All Commercial Service Vehicles
Using the Aggregate Demand Method

Travel Behavior Category	Description	Estimates
Fleet Size	Fleet size can be estimated as a function of population, based on data from truck populations.	0.05 per population (data from four cities).
Trip/Tour Length	Average mileages are consistent across different cities and categories, ranging from 29 to 49 miles per day. National average miles traveled will be derived from VIUS data. Average mileage will be derived from other commercial vehicle surveys.	41 average miles traveled per day, average trip length is 14 miles (data from eight cities).
Trips	Trips per vehicle can be derived from a commercial vehicle and government vehicle survey.	Three daily trips per vehicle (data from four cities).
Vehicle Miles Traveled	Service vehicles typically range from 5 percent to 13 percent of total VMT (based on estimates from other cities derived from DMV and VIUS data).	5.9 percent of total VMT (data from four cities).

Network-Based Quick Response Method

Data on public and private service vehicles were available for only four cities: Los Angeles, San Francisco, San Diego, and Sacramento. No data was available for the number of vehicle trips or mileages for these four cities because the DMV data for those cities contains only fleet size. Data on vehicle trips and mileages are available from commercial vehicle surveys for private sector service vehicles for the cities of Atlanta, Denver, Detroit, and the Triad cities. Additional data are necessary to more accurately evaluate travel behavior for all service vehicles. Table 4.4 presents a summary of the travel behavior characteristics for the Network-Based Quick Response Method.

Table 4.4 Travel Behavior Characteristics for All Commercial Service Vehicles
Using the Network-Based Quick Response Method

Travel Behavior Category	Description
Trips/Tours	Cross-classification or regression models can be used with employment variables. Government, utilities, and business and personal services employment are the most likely variables. Trip rates will be based on the truck population data and the Bureau of the Census. Typically, there are 0.1 per total employment or 0.05 per population.
Distribution	All service vehicles are distributed widely throughout the region and could be distributed with a gravity model. National average miles traveled will be derived from VIUS data. Average trip lengths will derived from other commercial vehicle surveys.
Vehicle Type	Service vehicles are primarily light-duty vehicles, dominated by public service, business, and personal service types (all light-duty vehicles). Some safety and utility vehicles are medium- and heavy-duty trucks (fire trucks, ambulances, utility trucks, etc.). Of all the commercial service vehicles, 91 percent are light-duty vehicles and 9 percent are medium-/heavy-duty trucks (based on data from other cities).
Time of Day	The majority of private service vehicles operate between 9:00 a.m. and 3:00 p.m., based on private service vehicles from the commercial vehicle surveys. The majority of public service vehicles also operate in this period. Of all total trips, 11 percent occur in the a.m. peak, 23 percent in the p.m. peak, 53 percent in midday, and 14 percent at night (data from other cities).
Trip Assignment	Service vehicles operate on all facilities.

4.2.3 Integration of Trucks in Four-Step Passenger Models

The truck-trip generation process in a four-step travel model system is independent of the passenger modeling components. The socioeconomic and demographic (SED) data is often shared between these two models that serve as the basic input providing a host of independent variables to compute productions and attractions. After the truck Ps and As are computed, they are fed into the truck distribution process which requires skim data that may include either travel time or distance. These skim data are derived from the assignment process which is a common modeling component for the truck model as well as the passenger model. This is the first point of integration between the truck model and the four-step passenger model. More details about this process are described in the trip distribution section of this section.

After the trip distribution models, truck trip tables are produced that are ready to be assigned to the highway network along with other modes considered in the passenger model. Trucks are much larger in size than the passenger cars and the presence of these large and low-performance vehicles in the traffic assignment process results in a reduction of the roadway capacity. The Highway Capacity Manual (HCM) cites that the reduction in roadway capacity is due to the fact that heavy vehicles such as trucks take up more space and have lower performance, especially on grades and during congestion. So the traffic volumes containing a mix of vehicle types of different sizes must be converted into an equivalent flow of passenger cars often referred to as the passenger car equivalents (PCE).

Different models use different PCE factors for trucks that are appropriate to the local region. It also depends on the different sizes and speeds of trucks in the model; the ideal way to calculate PCE factors is by collecting observed data. This can be done by gathering information on the vehicular composition at certain key segments of a region’s highway corridors that also includes speeds, travel times, grade, and congestion. As the data required for such an elaborate method is often scarce, most urban models assume these factors and calibrate them during the assignment process.

The PSRC truck model that includes three classes of trucks assumed light trucks to be equivalent to 1.5 passenger cars, medium trucks at 2.0, and heavy trucks at 2.5. After several rounds of calibration with more recent data, the PCE factors were updated and are now 1.0 for light, 1.5 for medium, and 2.0 for heavy trucks. Similarly, in the San Joaquin Valley truck model, there were no observed data available to support the development of PCE factors specific to the San Joaquin region. Therefore, the PCE factors used in the model based on guidelines provided by the Institute of Traffic Engineers were 2.0 for medium-heavy and 2.5 for heavy-heavy trucks.

The current SCAG HDT model includes a state-of-the-art PCE factor methodology that accounted for roadway grade, congestion levels, and percentage trucks in the traffic stream. The variable PCE factors have proven to be complex in their implementation and do not always represent the assignment process accurately. In the ongoing SCAG HDT model update, the variable PCE factor approach is being evaluated based on recent data to determine if it results in more accurate assignments. One area where the variable PCE factor does appear to provide improved assignments is the adjustment related to roadway grade. In the SCAG HDT model update, the locations where grade have been incorporated in the network are being reviewed for accuracy and additional locations with significant grade are being identified and incorporated in the highway network.

4.2.4 Data Requirement for Truck Models

In order to determine the data required to build a truck travel model, the first step is to assess the various truck parameters that need to be estimated. In statistical terms, these also are referred to as the dependent variables that depend on a host of explanatory or independent variables that often serve as the inputs to an urban truck model. The truck parameters of primary interest, but not limited to, are:

Truck productions and attractions by land use or sector or trip purpose;
Truck trips per day by truck type (GVW, FHWA class, etc.);
Truck trip lengths by truck type;
Truck trip time-of-day distributions; and
Truck volumes.

The aforementioned parameters are dependent on various inputs or independent variables that include, but not limited to:

SED Data or Employment Data – These data are essential to estimate truck production and attraction trip rates which are a function of observed truck trips coming into and going out of various land use types for which the SED or employment data are known beforehand. The observed truck trips are determined based on truck travel surveys. Different models use different types of employment data depending on the availability for the base and forecast years. Most of the current urban truck models use the two-digit SIC system of employment data. The level of aggregation or disaggregation of these into a finite number of categories depends on the variance of truck travel patterns associated with different land use types. The variance largely depends on the region’s economic activity that includes production and consumption of commercial goods. More recently, the NAICS system of employment data is being developed to better correlate and associate various employment categories to different types of businesses prevalent in an urban area.
Level of Service Data – These data include travel times and/or travel distances of vehicles in an urban area. This data is produced within a model system and is often known as the skim data. The skim data is an essential input to the gravity-based trip distribution models that estimates truck trip interchanges. The skim data is used as an independent variable to compute the travel impedances, which is then used to allocate the truck productions and attractions from the trip generation model to the appropriate origins and destinations in a region. This results in a truck trip table matrix, which is used in combination with truck travel distances to calculate the average truck trip lengths and frequency distributions.
Time-of-Day Factors – The truck travel surveys or classification counts are normally used to determine the time-of-day factors, which are proportions of truck trips occurring during a finite set of time periods. These time periods are decided beforehand depending on the level of detail necessary for an agency’s transportation planning purposes. The proportions or factors are applied to the daily trip tables coming out of the trip distribution model to produce trip tables by time period. These time period specific truck trip tables are then assigned to the traffic network along with the corresponding time period specific passenger trip tables.
Truck Classification Counts – The most important data that cannot be transferred or borrowed are the classification counts. Every model update includes the collection of these data. These are used to calibrate and validate the traffic assignment process that includes both passenger cars and trucks. Some agencies have a continuous traffic count program on key facilities such as freeways and expressways that are used in regular time intervals to update regional travel models. The level of detail of truck counts by various truck types or classes largely depends upon the truck model structure. Most count programs collect axle-based truck classification counts as these are easily captured by manual and machine counters. Agencies that use truck models based on GVW ratings convert the axle-based truck counts to appropriate GVW classes based on internally developed algorithms. The count locations also are important in the validation process of a truck model. These are usually collected on all the major facilities such freeways, expressways, and arterials. These also are collected at various points on a screenline and many screenlines are defined upfront of the count program. In addition to counts, other observed data that is necessary are truck speeds or travel times on key routes.
Level of Geography – Truck models are usually developed at the same level of geography as the passenger travel models. Almost all of the known urban area models use the TAZ-level geography. The primary reason for this being that all of the input data to a truck model is being developed at the TAZ level. There also are some aggregate levels of geography such as districts, super-districts, and counties that are often used to summarize truck model outputs during validation processes.
Roadway Networks – This forms the backbone of any model development effort that represents any region’s transportation infrastructure system. Truck models often use the highway networks that are developed for the passenger travel models and appropriate modifications are made based on the truck travel characteristics. These include coding truck only lanes, truck prohibition lanes, and/or truck priority lanes. The "truck" mode of travel also is coded as a separate mode to distinguish from other passenger travel modes and to determine truck travel volumes.

4.2.5 Special Generators at Intermodal Terminals

An intermodal terminal can be defined as a location for the transfer of freight from one transport mode to another such as between water and road (ports), road and rail (rail yards), or air and road (airports). The coordination of resources to achieve intermodal efficiency is a challenging task that involves government, the private sector, and various interest groups [http://www.doi.vic.gov.au/DOI/Internet/Freight.nsf/AllDocs/]. Intermodal terminals, which include seaports, airports, and rail terminals, serve as principal interchange points for both international and domestic freight movements.

The data collection efforts at intermodal terminals are always a challenge owing to the enormous time and costs associated. In addition, these data are specific to each type of intermodal terminal and cannot be transferred or borrowed. Specific models also are built based on the capacity and volume of traffic being handled at these facilities. The Southern California Association of Governments (SCAG) HDT model and Los Angeles Metropolitan Transportation Authority) LAMTA CubeCargo model are perhaps the only two models that capture the truck traffic coming out of and going into each of these three intermodal facilities in the region at the TAZ level.

Port Model

The port model for the SCAG HDT model included trip generation and distribution components. The port trip generation model was developed based on a detailed port area zone system and specialized trip generation rates for automobiles and trucks by type (Bobtail, Chassis, and Containers). The model generates three outputs – container terminal truck trips, container terminal automobile trips, and noncontainer truck trips. These three types of trips are usually the same across every seaport in the country. The Port of Long Beach (POLB) has a custom-built spreadsheet tool called the QuickTrip model that includes detailed input variables such as mode split (rail versus truck moves), time-of-day factoring, weekend moves, empty return factors, and other characteristics that affect the numbers of trucks through the gates. These factors vary by terminal at the ports, so a separate QuickTrip model is used for each terminal.

For trip distribution of port trips, a detailed and comprehensive truck-driver survey was undertaken at port marine container terminals. The surveys were used to develop detailed origin/destination “trip tables” for use in the port area travel demand model. The stated trip origin and destination from every valid survey was correlated with the travel demand model traffic analysis zone (TAZ) system. The survey results were then used to develop port truck origin/destination matrices by truck type for use in the model. The port trip matrices included a unique trip interchange percentage between every port marine container terminal and each of the model’s TAZs. This includes not only trips from marine terminals to land uses outside the ports, but also “interterminal” trips from one marine terminal to another marine terminal.

Rail Intermodal Facility

For LA MTA’s CubeCargo model, an innovative approach was used that yielded reliable information on the six rail facilities at a fraction of the investment in time and cost. The approach for the rail intermodal facilities began with contacts with the rail companies (BNSF and UP) regarding the six facilities. These contacts served a couple of purposes, namely, identifying the largest customers for each facility, and obtaining lift, gate, and train data. Additional data also was gathered that included lifts by day, split out by containers (international and domestic) versus trailers, and gate transactions by day by type (inbound, outbound, loaded, empty, bobtail). These data yielded the flow through the facilities without becoming entangled in short-term changes to train schedules and other operating adjustments. The train schedules themselves were available on-line and were supplemented with railroad records of actual arrivals and departures since some trains are run as extra or second sections.

By contacting the six facilities, relevant facility data were obtained that included a few relevant features of the rail facilities such as total acres, number of parking spaces, number of gates, number of employees/contractors, etc. The major customer contacts yielded the location and nature of their facilities, the location of their major markets or customer concentrations, and their pattern of truck trips between their facilities/markets and the six rail intermodal terminals in both directions, including empties and trips to obtain empties for loading. This information was then used to characterize and construct trip matrices for the nonport portions of truck traffic to and from the six rail intermodal facilities.

Air Cargo Trips

In the SCAG HDT model, air-cargo truck trips come from the agency’s Regional Airport Demand Allocation Model (RADAM) that was developed as a separate two-step process – airport trip generation and distribution. Three types of airport truck trips were accounted in this process. Heavy-duty trucks associated with airport operations such as maintenance, supplies, deliveries, and retail facility support comprised one category. Traffic between the five airports or with destination points outside of the airport area formed another element, while internal trips made by trucks within the airports formed the third component. The process of air-cargo trip generation involved the conversion of air cargo tonnage to truck trips, using the factors and relationships developed as part RADAM. The distribution was developed based on approximations of air cargo trip interchanges between airports and RADAM TAZs.

4.2.6 Constraints to Trip Generation

The general notion of building a trip table involves assuming that productions equal attractions. Depending on the availability of truck travel survey data, trip rates for a given sector or land use are either considered the same for production and attraction or they are estimated separately at each trip end. If the trip rates are assumed to be the same at both ends, then typically these are land use-based trip rates.

If data permits estimating two different rates for production and attraction, then these may be either employment- or land use-based trip rates. That is, the employment at that particular land use will drive the productions and/or attractions for any given sector. For example, “retail employment” in a TAZ can produce and attract trips that belong to the “mail/parcel” sector, if the supported by the data. If there are 200 “mail/parcel” expanded trips that are produced from a “retail” store, and if there are 300 “mail/parcel” expanded trips that are attracted to a “retail” store, then the production rate will be (200 trips/retail employee) and the attraction rate will be (300 trips/retail employee). These rates also can be estimated based on regression techniques where the dependent variables if the number of truck trips for a given sector and the independent variables are different types of employment. The coefficients associated with each employment variable are the trip rates. In other words, every sector (or trip purpose) will have a production rate and attraction rate for every type of land use (or employment) where trucks in that sector make stops at.

In the event of different productions and attractions, these will need to be balanced during trip distribution, so that the total number of trips originating from a given TAZ equal the number of trips destined to that particular TAZ.

4.2.7 Borrowed versus Survey-Based Truck Models

The borrowing of truck trip rates is a very common practice due to the lack of good survey data. This should, however, be done with caution. Almost one-half the urban truck models across the nation are based on the 1992 Phoenix metropolitan area truck model. The current QRFM recommends using the trip rates and gravity models from this model as a starting point, and then calibrating the parameters until they validate well with observed local count data. There are some limitations to this approach that needs to be understood well before borrowing truck parameters from other area models. The observed count data will serve well to validate the truck trip assignments but there will be no data for calibrating and validating trip generation and distribution models. That is, the precise estimates of total number of truck trips within each trip purpose or sector cannot be collected through a vehicle classification count program. Trip rates can be adjusted only after looking at the assignment results. Also, the average trip lengths and trip length frequency distributions can be calibrated only to approximate values and distributions borrowed from other area models.

The best way to estimate truck-model parameters is by collecting data through truck travel surveys. Different types of surveys such as trip dairy approach, establishment surveys, shipper/receiver surveys, and intercept surveys, provide different aspects of truck travel characteristics depending upon the type of business sector or trip purpose of trucks. The many benefits of using survey data are that:

Truck trip rates by sector or trip purpose can be estimated precisely as it will be calibrated and representative of the local truck travel behavior;
Observed data on average trip lengths and trip length frequency distributions can be used to calibrate/validate the trip distribution model;
Precise time-of-day factors can be derived from the observed survey data; and
Information on local issues also can be gathered from truck operators and drivers that could include commodity carried, qualitative data on what shippers and truckers see as their most difficult infrastructure problems (i.e., difficult intersections, bottlenecks, bridges, turning radii, road conditions, etc.), what most impacts their operation, etc.

The major limitation of truck travel surveys is the cost associated to conduct them especially since the response rates are well known to be very low. A considerable amount of resources and expertise is required to administer and conduct a successful truck travel survey.

4.2.8 Market Segmentation-Based Mode Split

The market segmentation-based method uses information from commodity flow data and base year mode split to forecast future mode split. It assumes that commodity and length of haul are good predictors of mode choice. The market segmentation method looks at the base year mode split by commodity and origin-destination pair and assumes that this reflects the relative service characteristics of available modes in these traffic lanes.

The basic assumption in this approach is that mode share for each commodity/O-D pair remains fixed in the future. But in the real world, the changes in the mix of commodities traded and the trading partners do affect overall mode share. So while using this approach, forecasters can do “what-if” scenarios by focusing on those markets (commodity/‌O-D) where modes actually compete to see if changes in modal characteristics could actually have a significant change on an overall mode share. The main data component here is the commodity flow data. When used for modal diversion analysis, the focus is primarily on intermodal cargoes and this can be determined from commodity flow data.

The following is presented as an example that explains the market-segmentation method when applied to modal diversion analysis for a state or a group of states. Using GIS tools or a routing network, the first step might be to determine a 500-mile radius from the centroid of each zone within the study area, where a zone might be defined as a county. O‑Ds farther than 500 miles from a zone usually represent O-D pairs where rail could compete with trucking. Now, using CFS or other national sources, those commodities need to be identified for lengths of haul greater than 500 miles where rail captures a known share of the market (e.g., 20 to 70 percent). The next step is to identify commodity/O-D pairs (at least one trip end in the study area) where rail is competitive but rail share is less than 50 percent. This will help in conducting what-if scenarios to see what impact would be if rail share could grow to 50 percent or 70 percent in all of the competitive markets. The changes or results of this modal diversion analysis can be seen in the total tonnage splits. It is always better to use national data to identify commodities for which truck and rail compete than to use study area commodity flow data. This is due to the fact that the lack of rail services may be limiting local markets and that is what requires change.

Pros

The advantages of the market-segmentation method are that:

It is simplistic in approach and in application;
The data is usually available for such an approach and is easy to process; and
It is reliable enough for modal networks and characteristics that do not change over time.

Cons

The limitations of this approach include:

Insensitiveness to policy impacts on mode choice;
Insensitiveness to implications of network investment strategies on mode choice; and
Assumption that modal characteristics remain constant over time when in reality there is a lot of variation.

4.2.9 Assignment Models

Traffic assignment is the last step in a travel-model system and there are a couple of broad ways to assign trucks to a roadway network. Truck assignments on highways could be either fixed or dynamic path assignment. In a fixed assignment, trucks are assigned to already existent fixed paths, whereas in a dynamic assignment, a computer program builds paths for the trucks. The key factors that go into the building of these paths, fixed or dynamic, are:

Infrastructure limitations (low bridges, bridge weight limits, speed limits, etc.) affect route choice.
Specific routings are usually selected as a function of cost, average travel time, the reliability of that travel time, and the general quality of service for the operators (safety, amenities, etc.). This happens after taking into account the limitations among available route choices.
The route may need to use specialized equipment or facilities, such as refrigerated terminal, or the cargo may be restricted from certain routes, like hazardous material/‌cargo prohibitions in tunnels.
The route may take competition among truck carriers or between modes into account (ship to certain intermediate destinations, less than truckload handling).
The operational characteristics of the network may be important, such as special truck routes, climbing lanes, or truck exclusions. The conditions probably vary by time of day in urban areas, which may affect the routing.
Highway routings or traffic assignment may be affected by all of the aforementioned factors, but only a few of them may be considered by the fixed path or dynamic models.

Fixed-Path Assignment

Fixed paths are provided by others, that is, paths already built are used. It may sometimes represent current routings of traffic or results of another dynamic assignment (e.g., ORNL routes for CFS flows). In fixed paths, if the network attributes change, either because of new facilities or congestion, there is no easy way to vary the paths. Also, the business decisions of carriers (which railroads work together) that are not easily modeled can be defined in these fixed paths.

The basic procedure in any assignment is to translate trip-table flows into link flows on a network and to use those link flows to determine system performance. The intermediate step used to make this translation is the information about the path or sequence of network facilities (links) used to travel from an origin to a destination. The basic feature of these paths in this assignment method is that they are fixed and would not vary, depending on network condition, congestion, new facilities, etc. These fixed paths can come from a variety of sources (TRANSEARCH, MapQuest, etc.). Invariably, these paths were created by the dynamic assignment methods that are described in the following section, but they have been saved by others as fixed paths for use in these assignments.

Once the paths are identified, it may be desirable to find the network flows a) for only selected origins or destinations (selected zone assignment); b) for only selected commodities (selected purpose assignment); or c) for only those flows that use certain facilities (selected link assignment).

For fixed-path assignment, the network needed does not have to be as rigorous as those used in sophisticated models, although for data management purposes alone, it is highly advisable. The paths are a file of the sequence of links used between each origin-destination pair. In order to produce system performance, the performance attributes for each link are required. These include information about the network and information about particular links in terms of travel times, costs, and distances among other parameters.

Fixed-path assignment methods are typically used to analyze long-haul traffic patterns at state or multistate level. Since the trips are over very long distances, the routing decisions are less responsive to local changes in network conditions and may remain fixed over long periods of time. These methods often develop deficiencies as traffic grows over time. However, they cannot be used to examine alternatives as the assignments are not responsive to network changes. Since routing models for nonhighway modes are generally proprietary or carrier-specific (and routing choices are more limited than for trucking), fixed path assignments can be very well used in these applications.

Dynamic Path Assignment

In dynamic assignment, paths are calculated by a computer program and may be used and discarded without the planner ever seeing them. Since the dynamic paths are computed as they are used, it is possible for the assignment to account for changes in the network. Dynamic assignment is the most commonly used process in urban automobile and transit passenger modeling. The outcomes from a dynamic assignment are similar to those of the fixed path assignment, such as link flow and network performance; however, dynamic path assignment can take congestion into consideration.

These paths also are a file of the sequence of links used between each origin-destination pair, but these files are temporary and created by the computer program. In order to calculate system performance, the impedance attributes are used to calculate the performance of each link. These include information about the network and information about particular links in terms of travel times, costs, and distances among other parameters.

Dynamic assignments can be used for any level of geography for which flows and networks are available and is the approach often used for modeling truck traffic at the metropolitan level. It is a more accurate way to estimate the impact of congestion on freight system performance, as the model can calculate new routes as congestion increases. It also is the best approach for alternatives analysis because the network can be modified to reflect alternative investment projects.

As explained for fixed-path assignments, the basic procedure in any assignment is to translate trip table flows into link flows on a network and to use those link flows to determine system performance. The intermediate step used to make this translation is the information about the path or sequence of network facilities (links) used to travel from an origin to a destination. In a dynamic path assignment, this path file is temporarily created within the assignment program. Just like in a fixed path assignment, the paths calculated in a dynamic process can be applied to perform selected zone assignment, selected purpose assignment, or selected link assignment.

For dynamic-path assignment, the network needs to follow the rules of the assignment program. In order to produce system performance, the performance attributes for each link are required and coded on to the highway network. These include information about the network and information about particular links in terms of travel times, costs, and distances among other parameters.

It is possible to calculate a wide variety of performance measures for dynamic-path assignments. It also is possible to do assignments for selected groups of commodities or other parameters analogous to trip purposes in passenger travel demand models. It is relatively complex to implement since special networks and software are required. Since it is so complex, the results of changes to the network may be counterintuitive or at least not obvious beforehand. However, it is very easy to modify the paths to account for new facilities or network conditions.

Type of Dynamic Assignments

There are a variety of methods to dynamically calculate paths which are described below.

All-or-Nothing or Preload Assignment – In the All-or-Nothing procedure, also referred to as preload, freight traffic is assigned to network without recalculating times or costs taking capacity constraints into consideration. It is appropriate for long-distance traffic flows where there may only be one desirable path anyway. Since a straight All-or-Nothing assignment typically loads too many trips onto the major facilities, a procedure to adjust the impedances for nonmajor segments is often applied.
Multiclass or Simultaneous Assignment – Truck trips are usually assigned together with the passenger vehicle model, because congestion has a significant impact on travel times experienced by trucks. If either nonfreight trucks or other vehicle trip tables are not available for congestion calculations, then they are preloaded onto the network using an All-or-Nothing procedure. Some agencies believe that trucks should be preloaded in all cases, because they do not believe that trucks, particularly larger less maneuverable trucks that may be operated by drivers not familiar with alternative routes, are as likely as automobiles to change their paths in response to congestion. Truck volumes are converted to Passenger Car Equivalents (PCE) to account for the fact that larger trucks take up more capacity and congestion for assignment of both trucks and passenger cars. This is explained in detail under Section 4.1.3.
Stochastic Assignment – In a stochastic or random assignment, all reasonable paths are used and are typically used in urban areas. It takes multiple paths in a network into consideration, and the user has control over how big a difference from the shortest path is reasonable. In any event, equal time/cost paths between the same O‑D pair will receive an equal share of the O-D flows.
User-Equilibrium Assignment – In equilibrium assignments, the travel times are recalculated based on delays associated with a loading and paths are recomputed and combined, such that all used paths have same travel time. This method is generally used in urban areas where there is a lot of congestion, and it takes network’s current capacity into consideration. Under equilibrium conditions, traffic arranges itself on congested networks in such a way that no individual trip maker can reduce his cost by switching routes. The equilibrium method attempts to find a solution where all used paths have the same travel time by iterating between All-or-Nothing traffic loadings and recalculating link impedances, such as travel time, based on the link volumes and capacity after each iteration. In fact, equilibrium is capacity restrained, since link times are recalculated based on capacity after each iteration. Capacity restrained assignment typically refers to those assignments where the user, not the computer, chooses how to proportion the flows from each iteration. For example, under equilibrium assignment, the computer calculates and may decide that equal times are achieved if 33 percent of the first assignment flows and 67 percent of the second assignment are used. Under capacity restrained assignment, the user may decide beforehand that 50 percent of each assignment is to be used.

4.3 State Freight Forecasting

4.3.1 Type of Model, Zone Structure, and Networks

Freight models in states that are geographically small and densely populated with adjoining urban areas, such as Connecticut and New Jersey, tend to take the form of urban truck models discussed in Section 4.1 above and will not be discussed further here. Freight models in larger states, particularly those with larger rural areas and/or large percentages of pass-through traffic, such as Indiana, Florida, and Wisconsin, forecast freight in “four-step” commodity models, are a principal focus of this section. Still other states, such as Virginia, Tennessee, and Georgia, follow the general form of commodity model, but use acquired commodity freight tables in lieu of forecasting those tables in the trip generation and trip distribution, and will be discussed in Section 5.0.

State “four-step” commodity models are truly multimodal. The modes considered in these models typically include truck, rail, water, and air, even though the assignment step may only address trucks, and sometimes rail. As multimodal commodity models, the flow unit is common to all modes, and is typically tons. These models tend to be calibrated from annual commodity flow tables and the forecasts in the first forecasting steps will be annual tons.

Freight forecasting models, as all models, should have boundaries such that they internalize most of the trips that will be subject to forecasting. In the case of passenger modeling, these boundaries can be set at the jurisdictional boundaries of the state. Internal freight traffic within a state is typically no more than 25 percent of the flow total, and the flow to, from, and through the state due to national traffic comprise the majority of the freight flows. In order to properly forecast this traffic, the geographical area covered by state freight models typically is most of the continental United States, if not all of North America. The inclusion of modes that primarily travel distances of over 500 miles, such as rail, water, and air also suggests that the freight modal boundary should be much greater than just the state boundary. States that have developed “four-step” commodity freight models typically already have developed detailed travel-demand model zones and networks within the state boundary. These models and zone systems have been extended by inclusion of national highway and rail networks.

4.3.2 Integration with Four-Step Passenger Models

There is value in being able to forecast freight flows, even when those forecasts are not integrated with passenger forecasting models. However, those states that have developed “four-step” commodity freight forecasting models have almost always had an existing passenger model. That passenger model has a zone structure and at least a highway network that can be used in developing commodity freight models. There is an additional reason for integrating freight and passenger model. At least for certain modes, always for trucks and passenger automobiles, and less often for freight and passenger rail, the modal networks are shared by passenger and freight vehicles and theses vehicles will interact in causing and being impacted congestion. There are several issues that must be addressed in integrating the passenger and freight models. The time period for passenger models is typically daily, while the time period for state freight models is typically annual. Before combining the forecasts, the freight flows are typically converted to daily flow units. The passenger and freight models can be kept separate through the trip generation, trip distribution, and mode split steps. However, the socioeconomic and transportation data used by these respective models should be the same. The tables of travel times covering the same areas should be the same for both models. The employment for the freight model may include more detailed industrial classifications, but the employment data and forecasts should be consistent with the employment and zone totals that are used in the passenger model. The freight and passenger models need to be combined in the modal-assignment step and that is when the vehicles will be combined. Therefore, the issues that will be discussed in later sections include converting the commodity freight flow units to vehicles and, for highway assignments, dealing with the issues of combining trucks and automobiles through the use of PCE, and in what order the trucks and automobiles should be assigned and interact.

4.3.3 Data Requirement for State Freight Models

For statewide freight models, data are needed to develop and specify the equation used in the various steps, and forecast adapt is needed in the same format to create freight flow forecasts. In a passenger forecast, the equations and relationships are developed from a household survey of travelers. In freight models, a commodity flow survey, typically either the publicly available Census Bureau’s CFS or the private commercially available TRANSEARCH data available from Global Insight. These tables tend to have limitations that must be overcome in using them to survey as freight surveys for model development. The CFS is publicly available only for 114 zones nationally, while TRANSEARCH is available for county zones, but the number of zones increases the purchase price. The challenge in the use of both models, either through additional processing of the CFS, or eventually through the FAF2 database derived in part from the CFS or through purchase of TRANSEARCH, is to develop zone structures that are detailed within the model study area, the state, and increasing less detailed at distances from the state model area. The state counties in TRANSEARCH led their zone structure to be used at the aggregate level to develop district relationships between freight flow and an economic variable, usually employment, which can then be applied to smaller units of geography. The commodity table typically has what is referred to as two-digit level of detail. Employment data are needed at an industry detail matching this freight commodity structure. Even the 40-50 commodities available provide data management and computational challenges and commodities carried forward are typically those that are the largest and most important to the study area. The associated employment must be available for those important commodities but may be aggregated to less detail matching the aggregated commodities. For example, printing may be included with all nondurable manufactured goods while food products would be retained as a separate category.

These commodity-flow surveys also provide information needed to calibrate the trip distribution and mode split steps. Commodity flows will typically need to be converted into units of daily vehicles because this more easily integrates with passenger forecasts and other transportation design, and operations tasks are typically based on daily flows. Data are needed to develop factors that can be used to convert from annual tons to daily trucks. The model needs to be validated to observed counts. This validation data, on highways, is observational, such as truck classification counts and typically will have no information on the commodities being carried. Since observational counts also include no information on truck purpose, those counts probably include trucks carrying local delivery of local freight or trucks used in construction, service, and utility trucks, none of which are included in the freight commodity model. Conversion from annual flows to daily modal vehicle flows is needed only for those modes that will be used in assignment.

In addition to calibration data, there is a need for forecast variables. The creation of a model that forecasts freight flows based on detailed industry employment for the zones in the model provides no value unless the detailed employment forecast can be obtained or created for the same industry and geographic detail in that same detail on coverage similar to zone structure.

4.3.4 Trip Generation

Trip-generation equations allow the development of forecasts for the flow of freight entering or leaving a zone based on economic conditions in that zone, most often employment. Since the amount of freight consumed or produced by employees will be different commodities and both in the amounts and the types of industries involved, these state models develop different equations for different commodities. The number and types of commodities to be included depends largely on the computational resources available and the economy of the state. These equations are developed through regression of the observed commodity survey data to employment by industry. Examples are provided in this section for the trip-generation equations developed for the Indiana, Florida, and Wisconsin “four-step” commodity freight models. Indiana developed trip generation equations using the 1997 CFS as the sample survey and employment by NAICS industry as the independent variable in the regression as shown in Table 4.5.

Table 4.5 Indiana Freight Model Variables used in Trip Generation

NAICS Employment	Description
212	Minerals and Ores
311	Food Manufacturing
312	Beverages and Tobacco
313	Textiles and Fabrics
314	Textile Mill Products
315	Apparel and Accessories
321	Wood Products
322	Paper
324	Printing
325	Chemicals
326	Plastics and Rubber Products
327	Nonmetallic Mineral Products
331	Primary Metal Products
332	Fabricated Metal Products
333	Machinery, Except Electrical
334	Computer and Electronic Parts
335	Electrical Equipment
336	Transportation Equipment
337	Furniture and Fixtures
421	Wholesale Trade, Durable Goods
422	Wholesale Trade, Nondurable Goods
POP	Population

Indiana developed equations for each of the two-digit Standard Classification of Transported Goods (SCTG) commodity categories used in the CFS. The production equations are shown in Table 4.6. In almost all instances in these equations, the employment variable in the production equation is related to the related industry producing the commodity. The equations produce annual thousands of tons of freight shipment by all modes. For example, according to the regression developed from the Indiana CFS data as shown in Table 4.6, each employee in the Chemical Industry (NAICS 324) produces 3,151 tons of Chemicals (SCTG 20) for shipment each year, with a “goodness of fit” (R-squared) of 78.2 percent.

Table 4.6 Indiana Freight Model Production Equations
Thousands of Annual Tons

SCTG	Name	Coefficient Times (NAICS3 Employment) [See Table 4.2]	Degrees of Freedom	R‑Squared
1	Live Animals and Fish	0.003(331)+.007(337)	22	0.498
2	Cereal Grains	0.256*(311)	36	0.337
3	Other Agricultural Products	0.135*(311)	34	0.647
4	Animal Feed	0.149*(311)	41	0.772
5	Meat, Fish, Seafood	0.054*(311)	42	0.880
6	Milled Grain Products	0.045(311)+0.027(333)	43	0.853
7	Fats and Oils	0.000748(Pop)+0.141(335)+0-083*(311)	46	0.964
8	Alcoholic Beverages	0.0002188(Pop)+0.013(334)	46	0.882
9	Tobacco Products	0.009(313)+0.005(337)	19	0.690
10	Building Stone	0.016(422)+0.0001118(Pop)+0.005*(331)	22	0.919
11	Natural Sands	0.087*(421)	28	0.839
12	Gravel and Crushed Stone	0.835(326)+1.145(314)+0.443*(311)	40	0.940
13	Nonmetallic Minerals	0.226*(325)	29	0.507
14	Metallic Ores	Not Applicable	[no data]	[no data]
15	Coal	7.34*(212)	30	0.604
17	Gasoline and Fuel	7.812*(324)	44	0.873
18	Fuel Oils	4.017*(324)	45	0.939
19	Products of Petroleum	3.388(324)+0.142(325)	41	0.918
20	Basic Chemicals	3.151*(324)	43	0.782
21	Pharmaceutical Products	0.011(337)+0.007(313)	35	0.793
22	Fertilizers	0.00081*(Pop)	35	0.304
23	Chemical Products	0.025(332)+0.017(325)	44	0.790
24	Plastics and Rubber	0.912*(324)	46	0.709
25	Logs and Rough Wood	0.667*(321)	21	0.518
26	Wood Products	0.544*(321)	44	0.826
27	Pulp Paper	0.225(322)+0.058(324)	44	0.810
28	Paper Products	0.029(311)+0.015(334)+0.053*(314)	45	0.931
29	Printed Products	0.024(422)+0.040(322)	43	0.946
30	Textiles and Leather	0.101(314)+0.051(313)+0.058*(324)	44	0.970
31	Nonmetallic Minerals	0.002(Pop)+0.248(311)	45	0.909
32	Base Metal	0.356(331)+0.080(336)	45	0.911
33	Fabricated Base Metal	0.030(332)+0.266(324)+0.033*(327)	45	0.949
34	Machinery	0.019(333)+0.026(326)	47	0.897
35	Electrical Equipment	0.017(332)+0.074(324)	46	0.913
36	Vehicles	0.061*(336)	44	0.798
37	Transportation Equipment	0.008*(331)	33	0.620
38	Precision Instruments	0.001*(421)	39	0.826
39	Furniture	0.020(337)+0.004(336)	45	0.918
40	Miscellaneous Manufacture	0.000183(Pop)+0.066(314)+0.022*(311)	39	0.946
41	Waste and Scrap	0.099*(332)	37	0.931
43	Mixed Freight	0.0004*(Pop)	38	0.905

The attraction equations are related to the industries that consume commodities. Although it is possible to test all possible employment by industry to determine the statistically most significant industries, that effort may be considerable. To assist in the development of these equations, candidate industries, as well as population for consumer goods that will be tested in the regression, are identified by examining national input-output models. Indiana developed equations for each of the SCTG two-digit commodity categories. The attraction equations are shown in Table 4.7. For example, according to the regression developed from the Indiana CFS data shown in Table 4.7, each employee in the Food Manufacturing Industry (NAICS 311) consumes 315 tons of Base Metal (SCTG 32) for shipment each year and each employee in the Transportation Equipment Industry (NAICS 336) consumes 79 tons, with a “goodness of fit” (R-squared) of 91.1 percent. It must be noted that it is not the point of this manual to justify these equations or relationships, nor to suggest that they are transferable to other regions, only to suggest that these are the findings for this freight model. It may be that these relationships indicate commodities being consumed that are locally prominent but not obvious unless more detailed information on commodity shipments, (i.e., shipment information for more digits in the hierarchical commodity classification system) is available. It also may be that the correlation is merely a spurious statistical aberration or a correlation with another more meaningful variable. Those developing the models should be aware of these concerns before choosing the variables to be used.

Table 4.7 Indiana Freight Model Attraction Equations
Thousands of Annual Tons

SCTG	Name	Coefficient Times (NAICS3 Employment) [See Table 4.2]	Degrees of Freedom	R‑Squared
1	Live Animals and Fish	0.004*(311)	18	0.488
2	Cereal Grains	2.724 *(324)	37	0.399
3	Other Agricultural Products	1.196*(324)	45	0.504
4	Animal Feed	0.148*(311)	45