Quick Response Freight Manual II

11.0 Applications Issues

11.1 Introduction

The purpose of this section is to give additional guidance on the application of the methods presented in this Manual. Many of theses applications issues are not unique to freight forecasting but are common to all forecasting, transportation or otherwise, but because they do affect freight forecasting, they are discussed here. Other application issues are unique to freight forecasting. Even though some of these issues have been discussed during the sections on the methods, for the sake of clarity they are discussed again in this section.

The earlier sections in this Manual discuss methods and data collection, but not all of the methods and data may be appropriate for a given area. When applying the methods of this manual, it is important to understand:

• What is the nature of the freight system in the area?
• What are the desired uses of the forecast?
• What is the availability and quality of data?
• What level of accuracy is needed, taking into consideration how the freight forecast relates to passenger forecasts?

Often, forecasts of freight are only a small part of a larger forecast encompassing both passenger and other types of truck travel. In these instances, the goals of the whole forecast need to be considered. Determining the level of effort should be based on an understanding of its importance in the whole forecast and to its potential contribution to the accuracy of the whole forecast. For example, if trucks comprise only 10 percent of the traffic in the area, then it would seem unreasonable to spend 50 percent of the forecasting effort on the freight portion. The challenge is to produce quality results by being resourceful, while still being efficient.

The issues of the nature and importance of the freight system, the uses of the forecasts and data availability have been discussed previously and will be discussed later in this section. The following discussion of level of accuracy is meant to reassure those who will be preparing forecasts of freight that the level of accuracy for freight forecasts need not always be held to the same standards as those of passenger forecasts. Error theory states that the most unreliable inputs have the greatest impact on the quality of the outputs. For example, on a major arterial in a city that has 20,000 vehicles per day, with 10 percent trucks and 90 percent passenger cars, the root mean square error (RMSE) of assigned passenger car volumes might be 15 percent, or 2,700 cars (15 percent = 2,700/[0.9*20,000]). If the error in the assigned volumes for trucks were to be 30 percent, then it would result in only a small increase in the totals error, making the typical assumption that the errors are independently distributed random variables.

• Total Error with a 15 Percent Truck Error = (2,700² + 270²)^½ = 2,713
• Total Error with a 30 Percent Truck Error = (2,700² + 540²)^½ = 2,753

The total RMSE error of all of the forecast volumes increases only 2 percent with the increase in truck error from 15 to 30 percent. Therefore, the truck forecast in this example can tolerate a much greater error than the passenger car forecast without adversely affecting the total vehicle forecast. Of course, this conclusion does not apply to instances where the freight forecast is of primary importance.

With the understanding that the level of accuracy in freight forecasting need not be as strict as those required for passenger forecasting, the remainder of this section will discuss model application issues from the previous sections.

11.2 Controlling Factors

Understanding the controlling factors for freight forecasting and how a particular study or project can influence and be influenced by these factors is crucial in freight forecasting. The forecasting process must be able to address those factors that are significant, address the major characteristics of the alternatives that are under consideration, and consider factors that are within the jurisdiction of the agencies for which the forecasts are prepared.

Shipment sizes and frequency may be an important factor in the choice of mode by shippers. These business process decisions may vary markedly between industries, are subject to continuous and unpredictable change, may not be within the jurisdiction of public agencies, and would be unlikely to be addressed by public agency projects or programs. Therefore, it may not be necessary or appropriate when considering mode split in the forecasting of freight to include variables like shipment size in the forecasting methods.

Different modes may only be appropriate when the shipping distances exceed several hundreds of miles. If the planning jurisdiction is a metropolitan area that cover a radius of less than 100 miles, either those modes may need to be handled separately or be excluded from the agency forecasting methods; or if the actions of the agency are likely to impact mode choice for freight, the forecasting process for freight may need to cover a broader area than that of passenger forecasting. Covering a broader area has implications on the development of a network and obtaining data and forecast for this larger area.

Management systems may require forecasts for a time period that is much shorter than typical 20-year horizon considered by planning agencies. Pavement management systems may require forecasts of freight volumes that cover a period of only several years and may need to be sensitive to seasonal variations that are not typically considered in that agencies forecasting process. In those cases, simple growth factoring methods may be better able to provide short-term forecasts that are sensitive to seasonal demand.

These are only a few examples of some of the ways that an understanding of the controlling factors can influence the selection of a freight forecasting method. If the effort to consider these factors is not made, an agency runs the risk of expending scarce resources in developing a method that cannot consider certain policy and project attributes or developing a method that considers attributes that will never be changed or for which forecast of those attributes will not be available.

11.3 Growth Factoring

Whether growth factoring methods rely on trends in historical flows or economic indicators, the use of growth factors assumes that the trend that existed in the past will continue into the future. More sophisticated forecasting methods should be considered when it is known that this assumption is not correct. One of the basic tenets in growth factoring is that if there is no activity in the past, then applying a growth factor to that lack of activity also will show no activity. In other words, zero times a factor will still be zero. In situations where new freight activity is expected not just an extension of existing trends, growth factoring is probably not appropriate. Similarly, if an underlying change in the freight activity is expected, that was not present during the period from which the growth factors were developed, then growth factoring may not be appropriate. An example of this may be growth factors that could be developed for a period that reflected the economic regulation of carriers, followed by a period without economic regulation. The use of factors based on these past trends into the future may not yield accurate forecasts.

In addition to the application issues associated with extending past trends, the use of economic indicator variables has additional issues. Developing the relationship between the freight flow and the economic indicator variables may be difficult. Truck counts may be available only in the aggregate and provide no information about underlying purposes or commodities that could be associated with economic indicators. Total truck volumes might be known, but flows by commodity might be unknown or difficult to obtain and average usage assumptions may not be appropriate. Average statewide truck usage assumptions by commodity, for example from VIUS, may not be appropriate for a specific corridor where the economic activity is different than the average of the whole state. For example, a corridor that has a higher average concentration of high-tech or service industries than the state average would not be expected to match average state freight flow patterns.

The establishment of a suitable geographic area for the economic indicator variables that can be associated with a facility may be difficult. If a facility carriers a great deal of through freight, then local employment might be a poor indicator of future growth. Even the establishment of a suitable influence area for a specific facility is difficult. This raises several questions, such as: Should it be surrounding census tracts? The county in which it is located? Surrounding counties? Substate areas? Even if the area can be determined, obtaining base and forecast economic indicators may not be possible. It is not possible to develop factors that are based on the growth in employment in a particular industrial sector when no forecasts of employment in that sector are available.

11.4 Network and Zone Structure

The ability to use four-step methods to forecast freight will be dictated by the network and zone structure that is available to support that analysis. The geographic area covered by the model may be too small to address the distances or cover the markets that need to be considered in freight forecasting. This is true whether the freight trip table is one that is created through a trip generation/trip distribution/mode split process or is one based on a commodity table obtained from other sources. The area covered by the model needs to cover the area which is expected to influence freight decisions.

Once the area to be covered by the model has been identified, the application issue will be to obtain base- and forecast-year data for that zone structure. In many cases, the zone structure at which that data are available for the geography outside of the model area will dictate the zone structure. For example, for a freight model for New York, if data is available only for the State of Florida, unless the forecasting process is going to estimate data at smaller levels of geography within Florida, then a single zone covering the State of Florida would be appropriate for this forecasting application.

The difficulty of obtaining networks outside of the area covered by the passenger travel model, i.e., state or urban area, and providing linkages between that model and the larger network is simpler than it has been in the past. FHWA developed a highway network in TransCAD format as part of its FAF1 project [FAF1 TransCAD network available for download from http://ops.fhwa.dot.gov/freight/freight_analysis/faf/faf_highwaycap.htm] and will soon be providing an update of that network as part of its FAF2 project. These networks have all of the attributes needed for travel demand model network (e.g., connected links and nodes, centroid connectors to county zones, link distances, functional classifications/facility types, capacities, free flow speed, congested speeds, etc.). The node and link locations are coded in a decimal degree projection and with sufficient detail (e.g., county FIPS code) to allow this highway network to be integrated with existing travel demand models covering areas of smaller geography. For other modes (i.e., rail and inland water), the BTS web site provides networks that can be downloaded and modified for use in travel demand models. [The National Transportation Atlas Databases 2006 (NTAD2006) is a set of nationwide geographic databases of transportation facilities, transportation networks, and associated infrastructure. These datasets include spatial information for transportation modal networks and intermodal terminals, as well as the related attribute information for these features. Available on CD or for download at http://www.bts.gov/publications/national_transportation_atlas_database/2006/.]

11.5 Trip Generation

Trip generation application issues will vary depending on whether the model is a commercial truck model, like those used in most urban areas or densely populated states such as Connecticut or New Jersey, or is a multimodal commodity model like those used in states such as Florida, Wisconsin, or Indiana. In the case of truck models, the trip generation will forecast trips by truck type (e.g., medium and heavy, single unit and combinations, etc.). For multimodal commodity models, trip generation will be for groups of similar commodities. In both types of models, the trip generation equations may be created by regression of the independent variable (most often employment by industry) to a survey of the dependent variable, observed base year flows (most likely a commercial vehicle survey) for truck type models, and a CFS for multimodal commodity models. If the rates are borrowed, then they still would have been created in this fashion in the area from which it is borrowed. In the event that the rates are borrowed, it should be recognized that the assumption is that the conditions giving rise to those trip generation equations also are similar enough to make borrowing appropriate.

Whether the equations are borrowed or created from a survey, the equations should have no constant terms. No economic activity means there should be no freight activity. For models where the trip purposes are truck trips, the production equations will be the same as the attraction equations, i.e., the number of trucks entering a zone should equal the number of trucks leaving that zone. For models where the purpose is a commodity, there should be different equations for productions and attractions, i.e., there is no reason for the flow of a commodity from a zone to equal the flow of the commodity to that zone. Additionally, the independent variable in the commodity production equation will likely be related to the industry producing that commodity, while the attraction industry will be related to the industries consuming that commodity.

For commodity models that are based on surveys of unlinked trips, that is where the survey includes a separate record for each modal portion of multimodal trip, the change of mode will not be able to be explicitly calculated in the forecast. In those instances, the traffic originating or destined for zones that contain intermodal terminals will be unrelated to economic activity in that zone for the producing or consuming industries. In the case of those commodity surveys that include only the domestic portion of an international shipment of freight, such as the CFS, the freight shipments to or from those zones containing international marine ports also will be unrelated to economic activity in those zones. These zones will need to be handled as special generator zones in the trip generation process. Forecasts for these special generators should ideally be obtained from other sources, such as the facility operators.

11.6 Trip Distribution

The exchange of freight between zones is limited by the total production and attraction of freight trips from the trip generation step and is governed by the impedance or friction to travel between zones, in a manner similar to that used in passenger forecasting. Almost all freight trip distribution methods use a form of the gravity model. In both urban and long-distance freight modeling, an exponential form of the impedance function is most often used. The use of the exponential form of the impedance function provides a useful check on the coefficients that are used in freight trip distribution. When the impedance or friction factor between zones i and j is of the form:

where the k‑coefficient in that exponential distribution is by definition the inverse of the average trip length expressed in the travel units, usually time or distance, measured by tij. Thus, in an urban truck model, when the travel unit is minutes and the k‑coefficient for a practical truck purpose is 0.08, the implied average travel time is 12.5 minutes (12.5 = 1/.08). For long-distance freight models, the average trip length is typically given; for example, a 526 miles mean shipping distance for metallic ores will have a the coefficient of -0.0019 (1/526). An average distance of 562 miles, typically a longer distance than can be traveled in most statewide models, shows why these models need to include national networks, well beyond the study area focus, simply to forecast the behavior of freight in the study area.

In passenger forecasting, the result of the trip distribution process is a production attraction table which must be later transposed into an origin destination table. In passenger modeling, this is due to the need to make sure that the trips between the origin and destination are balanced for trips based at the home, i.e., trips made from a home zone to a work zone must return to the same home zone at the end of the day. There is no need to account for this process in freight modeling. The result of a freight trip distribution already is in an origin destination format. A shipment of metal products from a factory will not return to the factory but will be consumed by producing other goods, and those goods will be forecasted separately.

11.7 Mode Choice

Unlike passenger forecasting, mode choice is not often addressed by an equation in freight forecasting methods. In truck-based models, there is no need to calculate the mode because, by definition, it is truck. In commodity-based models, additional research is needed to better define the utility variables that give rise to modal choice, as well as to develop credible estimates of these utility variables for modes other than by highway. In the absence of these methods, most commodity-based models rely on the underlying distribution of freight in the base commodity survey and assume that this mode choice, by commodity, will remain constant in the future. Since the approach relies on the base CFS, the approach is transferable, but the mode shares will be specific to the region’s CFS and are not transferable.

11.8 Conversion to Vehicles

For truck-based freight forecasting, there is obviously no need to convert to vehicles, since the flow unit already is expressed in vehicles. This step only is necessary for commodity-based multimodal freight forecasts. Generally, those forecasts will be calculated in a non-mode-specific flow unit, such as tons, ton-miles, or value. In order to be useful in many transportation forecasting applications, it is necessary to convert those flow units from annual tons to daily vehicles: most often trucks or less often rail cars.

A common source of information used to convert from tons to trucks is the U.S. Census Bureau’s VIUS. This survey includes records of truck usage, in terms of percentage of miles traveled carrying certain categories of cargo. The state records in this database can be used to develop average payloads from the weight of the vehicle surveys and the percentage of miles that it carries specific commodities. Prior to 2002, VIUS used its own unique commodity classification system, and used coding roughly equivalent to the SCTG commodity classification.

The vehicle payloads by commodity shown in this Manual can be transferred for use elsewhere, but it should be understood that the estimates are based on trucks based in that state. VIUS cannot be used to determine information for trucks traveling to a state nor for trucks traveling through a state. The payload mix for a state is based on the survey mix of commodities for trucks based in that state. Although truck characteristics can be expected to be similar everywhere, transferred rates should be used with caution. The VIUS includes information on the percentage of miles a truck is driven empty. Therefore, the VIUS-derived payloads can include allowances for empty miles. The Florida and Wisconsin values shown in Section 5.0 of this Manual include allowances for empty miles.

VIUS has not been funded as part of the 2007 Economic Survey. To the extent that the commodity carrying characteristics of freight are not expected to change over time, it may be appropriate to use the 2002 VIUS, which may be the last such survey undertaken.

The conversion from annual tons to daily tons also is a consideration that must be considered in converting to vehicle trips. This conversion will be based on local considerations on how an average day is included in other transportation forecasting. Typically, this number is based on the number of working days per year during which freight is expected to move. Values commonly used are 312 days per year (6 days per week), 306 days per year (6 days per week less 6 major holidays), or 250 days per year (5 days per week). This consideration also is where adjustments to reflect seasonal variations could be made.

11.9 Assignment

The results of the truck freight assignments in highway models can take one of two forms: 1) truck trips that will be pre-assigned to links before the passenger auto trips are assigned; or 2) a truck origin-destination trip table that will be assigned to the network at the same time as passenger auto trips. Depending upon the chosen assignment method and features of the software, each form has its advantages and disadvantages.

There may be valid conceptual considerations for pre-assigning assigning truck trips. The drivers of large trucks passing through an area may be unfamiliar with the possible alternate paths available in the event of congestion and may choose only the preferred paths. Large trucks may not be able to maneuver on the arterial and collector roads that comprise the alternate paths. Large truck companies/drivers may value reliability more than travel time and chose the certain travel time on congested routes over the less reliable time on faster alternate routes. There also are computational advantages of pre-assigning truck trips: 1) PCE factors can be adjusted for grade and other road conditions specific to individual links; and 2) certain links and turn movements can be prohibited. When trucks are pre-assigned, their volumes contribute to the congestion calculations for auto travel.

Assigning truck trip tables together with passenger auto trip tables in a multiclass assignment is appropriate when it is expected that trucks will respond to congestion in a manner similar to autos. This may be because the majority of truck drivers are familiar with alternate paths or congestion introduces unreliable conditions rendering all paths suitable for trucks. It is still possible in these multiclass assignments to restrict trucks or autos from certain links. It is just that both trucks and autos will modify their paths on all links in response to congestion. The computational advantages of assigning a truck trip table at the same time as a passenger car trip table are: 1) faster software execution; 2) less data manipulation; and 3) the ability to reroute trucks to avoid congested links and turns.

Assigning vehicles for modes other than trucks is not typically undertaken. When it is done, the assignment may use a predetermined set of paths between an origin and a destination that will not vary due to congestion. This approach is often the approach taken in forecasting rail flows on a network. The advantages of using a predetermined set of paths is that it can consider the private business decisions of the modal operators, where paths may be chosen to balance loads, maximize system revenue, provide incentives to favored shippers, or other reasons that may not optimize the paths for a specific shipment (user). However, these paths are usually obtained qualitatively through examination rather than quantitatively based on characteristics of the links on the path. Since the paths are not based on link characteristics, they cannot easily be changed in response to establishing new characteristics along the links, e.g., improved track speeds.

11.10 Integration with Passenger Forecasts

There are several application issues to consider when integrating freight forecasts with passengers models. Most of those issues occur in integrating truck forecasts into the highway model while the trip generation/trip distribution/mode split steps will remain separate from passenger forecast because they are being treated as different purposes. There is one purpose in particular in truck models that needs special consideration in integrating with the freight forecasting methods described in this manual. Four-tire trucks, FHWA vehicle Class 4, the category of vehicles that includes pickups, light vans, and SUVs, includes vehicles that can be used for both passenger and commercial purposes. The behavior of these light trucks in commercial purposes is very similar to that of all passenger vehicles used for Non-Home-Based (NHB) passenger trips. Those preparing forecasts will need to decide if these commercial trips should be considered separately or with NHB trips. Even if they are considered separately, the validation data available for the assignment of light trucks, observed truck counts on links, will not distinguish between passenger and commercial purposes of these trips and other means may be needed to validate these commercial light truck trips.

A similar application issue exists for commodity freight truck models in terms of how to integrate commodity trucks with passenger auto forecasts. In these models, the definition of freight may exclude local deliveries of freight, and those local delivery truck trips would not be included in the forecasts. The validation data that will be available, observed three-axle or higher truck trips on links, will not distinguish between trucks used in freight and trucks used in other purposes, such as for service, construction, or utility purposes, much less between the local delivery and commodity purposes of trucks. Validation of the truck portion of commodity models may need to be based on flows on links where these flows predominate over the local or non freight purposes; for example, on rural interstates and principal arterials. In this event, before the commodity truck forecasts can be integrated with passenger auto forecasts, some estimate of the remaining portion of commercial truck trips must be made. In developing noncommodity truck trip forecasts, for example, using the methods outlined in Section 4.1, it should be noted that these methods include commodity trucks and some means to exclude this portion of truck trips must be developed.