Forecasting Statewide Freight Toolkit (2008)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

9This Toolkit is organized on the basis of five basic model classes and six modeling components. The model classes share many basic components, as shown in Table 4.1. All of the classes, except the direct facility flow factoring model class, assign one or more modal tables to modal networks. The origin-destination (O-D) factoring, four-step commod- ity, and economic activity models all have mode split com- ponents. The truck, four-step commodity, and economic activity model classes all have trip generation and trip distri- bution components. The truck and four-step models use exogenously supplied zonal employment or economic activ- ity in the trip generation component, while the economic activity model forecasts the employment or economic activ- ity based on economic and land use data. Section 4.6 discusses economic activity/land use models, which, depending on the model class, can be integrated into the freight forecasting process, run separately to provide socio- economic data forecasts, or used to obtain growth factors. 4.1 Direct Factoring As shown in Figure 4.1, the direct factoring model compo- nent produces forecasts of link volumes, such as those on roads, railroad tracks, or ports, using basic information about existing flows and forecasts of economic data or trends that would affect the facility. This method uses existing freight flow for a facility, modal network link, or terminal. Factors are developed and applied to estimate changes in this facility flow due to growth or changes in transportation service on that facility or on a com- peting facility regardless of mode. Direct factoring is used in many states. Usually intended for short-term forecasts, the model component involves simple methods intended for rapid application of existing data to determine one or more forecasted items. Successful direct fac- toring requires many assumptions and the model’s range of applicability is limited. The Federal Highway Administration’s Guidebook on Statewide Travel Forecasting discusses time series methods for direct forecasts of vehicular volumes on highway and for forecasting the inputs to four-step models.5 The Guidebook emphasizes autoregressive integrated moving average (ARIMA) models and growth factor methods, while describing a linear regression model to forecast truck vol- umes on Interstate 40 in New Mexico. The Federal Highway Administration’s Quick Response Freight Manual describes two methods of applying factors to traffic volumes applicable to rural highways as well as urban highways.3 The first method involves estimating a growth factor from current and past truck count data and applying the resulting factor to future years using a conventional compound interest formula. The second method determines separate growth factors for vari- ous “economic indicator variables,” usually employment in local industrial sectors. The future growth in economic indi- cator variables, as calculated by a compound interest formula, is used to forecast growth in commodity groups. NCHRP Report 260: Application of Statewide Freight Demand Forecasting Techniques, describes a generalized pro- cedure of O-D table factoring and assignment.6 The report assumes that commodity production is directly related to employment in industries that produce the commodity. For estimating consumption, it recommends the use of an input- output table. Commodity consumption calculations follow a three-step process: 1. Obtain an input-output table; 2. Convert dollar amounts to tons and sum the columns of the table to find consumption by industry; and 3. Allocate tons to counties (the assumed size of the Traffic Analysis Zone, or TAZ) according to the employment by consuming industries and population (for final demand) in each county. These steps assume that the production and consump- tion estimates can be applied to an existing commodity C H A P T E R 4 Forecasting Components

flow matrix or (in the absence of a matrix) incorporated into a gravity model of shipment distribution. 4.2 Trip Generation As shown in Figure 4.2, the trip generation model compo- nent forecasts the productions and attractions of freight movements that begin or end in a geographic zone based on the characteristics of that zone. The most common charac- teristic used in trip generation is the employment by industry that produces and consumes various goods. The output of 10 trip generation, a production and attraction file for all geo- graphic zones, customarily serves as input to other model components used in freight forecasting. However, the pro- duction and attraction file can be useful on its own, showing freight trips that end in zones. The trip generation models used in statewide freight fore- casting include a set of annual or daily trip generation rates or equations by commodity, providing annual or daily flows originating or terminating in geographic zones as functions of TAZ or county population and disaggregated employment data. Production and consumption tonnages for special gen- erators like seaports, airports, and other intermodal transfer terminals are directly obtained from the port or terminal for the base year. The commodity flow tonnages for external zones are obtained from the commodity flow database and are disaggregated at the TAZ or county level based on the dis- tribution of employment within each TAZ or county. For the truck model class of freight models, trip generation is usually calculated separately for internal trips between zones (I-I) and external trips between internal and external zones (E-I, I-E, and E-E). Trip rates are derived from national sources such as the Quick Response Freight Manual and/or regional sources, if available. These are applied to households and employment data to obtain truck trips internal to the state. Different trip rates by truck type are used for truck trip productions and attractions. The socioeconomic data used in a typical truck model are consistent with those data used in Model Component Model Class Direct Factoring Trip Generation Trip Distribution Mode Split Traffic Assignment Economic/Land Use Modeling Direct Facility Flow Factoring Method Of facility flows O-D Factoring Method Of O-D tables Included Included Truck Model Based on exo- genously supplied zonal activity Included Not Applicable Included Four-Step Commodity Model Based on exo- genously supplied zonal activity Included Included Included Economic Activity Model Based on out- puts of eco- nomic model Included Included Included Included Table 4.1. Freight model classes by component. Data Direct Facility Flowing Factoring Link Volumes 1 5 Figure 4.1. Direct factoring.

passenger models, except that the employment data are strat- ified into more employment categories. This process provides more accuracy for truck travel, allows for a direct relationship between the commodities being estimated in the external trip model that captures truck flows in and out of the state, and helps allocate these commodities to traffic analysis zones within the state. The stratification of employment data is usu- ally by Standard Industrial Classification (SIC) codes at the two-digit level. The trip rates are usually adjusted during model calibration based on local or regional knowledge of truck trip ends. Section 5.6 includes a detailed discussion of SIC codes, and Table A.2 in Appendix A shows the corre- spondence between STCC codes and the Standard Classifica- tion of Transported Goods SCTG. The external truck trips (E-I and I-E) entering and leaving the state are derived from observed data at external stations and truck survey data. These data are disaggregated to the TAZ level based on percent distribution of various employ- ments by industry within each internal TAZ. Some truck models also use commodity flows that have either their ori- gin or destination within the state boundary. This process involves the conversion of commodity flow tonnage to truck trips. The through trips (E-E) that pass through the state with both origin and destination outside the state are added to the external truck trips as well. Trip rates are applied to socioeconomic data and also are used for truck terminals and intermodal facilities in conjunc- tion with observed truck trips at airports, seaports, and rail terminals. The commodity-based trip generation model includes a set of annual or daily trip generation rates or equations by com- modity, providing annual or daily flows as functions of TAZ or county population and disaggregated employment data. The Florida Intermodal Statewide Highway Freight Model, described in Section 8.9, uses Reebie Associates’ TRANSEARCH freight database to derive linear production and consumption equations for 14 commodity groups. The independent variables are primarily population and employ- ment by SIC at the county level for the State of Florida. The employment categories are based on the U.S. Department of Commerce’s Bureau of Economic Analysis 1996 input-output tables and tailored to the commodity group being estimated. Production and consumption tonnages for special genera- tors like seaports, airports, and other intermodal transfer terminals are obtained directly from the port or terminal for the base year. The commodity flow tonnages for external zones are obtained from the commodity flow database and are disaggregated at the TAZ or county level based on the dis- tribution of employment within each TAZ or county. As shown in the Section 8.8 case study, the Indiana Com- modity Transport Model includes 21 commodity groups considered important to the state. The trip generation equa- tions were developed based on a regression of data available from the Bureau of Transportation Statistics’ 1993 Com- modity Flow Survey (CFS). The Nebraska Statewide Freight Forecasting Model also uses the 1993 CFS data to develop a trip production model. However, IMPLAN software pro- vided input-output coefficients that were used to derive trip attraction equations. The Vermont Statewide Freight Study uses O-D data from the TRANSEARCH database, organized at the two-digit STCC level to build the trip tables. In addi- tion to commodity flows, the Vermont study uses roadside surveys, motor carrier surveys, and data from interviews with key shippers to develop the trip tables. The Iowa State- wide Freight Transportation Model also uses the Reebie TRANSEARCH commodity data, organized by Bureau of Economic Analysis zones at the two-digit STCC level. The Nebraska, Vermont, and Iowa models are not the subject of case studies in Section 8.0, but are cited in the References sec- tion of the Toolkit. Section 5.6 describes STCC codes in greater detail. 4.3 Trip Distribution As shown in Figure 4.3, the trip distribution model com- ponent produces the production and attraction file for zones to forecast a table of freight flows between all geographic zones. The trip distribution model also requires some infor- mation about the degree of difficulty for freight to travel between all zones. With the exception of single mode models 11 Data Trip Generation Multimodal P & A File 2 1 Figure 4.2. Trip generation.

such as the freight truck model, the flow in the trip table is expressed in units that are common to all modes. When the freight trip table is a multimodal commodity table, it cus- tomarily serves as input to the mode split model component. When the table is for a single mode it customarily serves as input to the assignment model component. However, the trip tables themselves are useful in analyzing the markets for freight flow between geographic zones. The trip distribution models are used in statewide models to forecast the volume of freight shipped between an origin and a destination. All the state freight models surveyed use gravity models for distribution. Gravity models distribute trips by purpose between origins and destinations, based on the total tons produced at an origin, attracted to a destina- tion, and the relative impedance, in the form of friction factors, of traveling between these zones. Gravity models cal- culate this distribution for each O-D pair by purposes and adjust the calculations iteratively based on the calculations of all other pairs of the same trip purpose. Truck models use truck types as trip purposes. For the New Jersey Statewide Truck Model described in Section 8.6, the trip purposes were light, medium, and heavy trucks. These were distributed from origins to destinations using the gravity model technique, the same method used in any typical automobile passenger model. The friction factor curves are first derived from the Quick Response Freight Manual and later adjusted to provide the best fit with average trip lengths derived from observed truck survey data. The friction factors were developed using the following equations from the manual: Light = exp (−0.08 * congested travel time) Medium = exp (−0.10 * congested travel time) Heavy = exp (−0.03 * congested travel time) The New Jersey model employs different gravity models for internal, external, and through-trips for both medium and heavy truck types. These models are calibrated to match target distributions based on a combination of observed data for trips in New Jersey where data are available, and data from other cities where local data are unavailable. The commodity-based freight models use gravity models for trip distribution. However, rather than being trip-specific, the models are developed and applied for commodity groups serving as the purposes for individual tables. Freight flows in tonnage and by commodity group are distributed on an O-D basis for an entire state, either at a district, county, or TAZ level. The primary impedance variables are average travel dis- tance, average travel time, or composite modal travel time. The trip distribution component for the Florida Inter- modal Statewide Highway Freight Model described in Section 8.9 uses a standard gravity model and distributes tons produced in one zone to tons consumed in another zone using friction factors calibrated based on the average trip lengths identified from TRANSEARCH. In the Indiana Commodity Transport Model described in Section 8.8, freight shipments are distributed by a gravity model cali- brated using the CFS data. Special care is taken to match the average shipping distance per ton for each commodity group. This prevents any inappropriate weighting for many short-distance lightweight deliveries versus a few long- distance heavyweight shipments that might be included in the same commodity group. 4.4 Mode Split As shown in Figure 4.4, the mode split model component uses a freight trip table, obtained either from the trip distri- bution or the commodity flow model components, to fore- cast tables of freight flows between all geographic zones for individual freight modes. The mode split component also requires some information about the relative benefits of the utility of using each freight mode between all geographic zones. The modal trip tables of freight flow customarily serve as inputs to the assignment model component. If the flow is not expressed as vehicles, but in flow units common for all modes such as tons, a conversion to vehicles may be made prior to using the tables in assignment. However, the trip tables themselves are useful in analyzing the markets for freight flow between geographic zones by mode. 12 Trip Distribution Multimodal Matrix Multimodal P & A File 3 2 Figure 4.3. Trip distribution.

Mode split models, as they are used in statewide freight forecasting, convert future flows by all modes into flows by specific modes. (By definition the truck model class involves only a single mode.) A mode split model may use modal shares from the base year commodity data by origin, destina- tion, and commodity group to determine the mode split in the forecast year. These are usually not sensitive to factors like travel times, travel costs, safety, and reliability. By identifying specific markets that may have the option to switch modes based on the distance traveled, the type of commodity, and the size of the shipment, it may be possible to qualitatively adjust mode shares. If modal utility data is available, that information can be used together with the base commodity flow data to develop freight mode split models. Commodity flow tonnage is converted to vehicles based on commodity-specific factors (tons per truck or railcar) so that loaded truck and/or railcar trips can be assigned to the corre- sponding networks. In most of the models, conversion to air and waterborne vehicles is not undertaken since assignment to air and water networks is typically not performed. How- ever, the Texas statewide model does convert barge traffic to waterborne tons. For the O-D factoring class of models, in the process of fac- toring existing O-D commodity tables by modes, each exist- ing modal table is often factored separately. Implicitly, this assumes that existing mode shares for each commodity will continue in the future. This is not the only option for treat- ing the split into modes within the O-D factoring class of models. How modal allocation is treated depends on the modal-specific network information that is available. If no information is available on the travel times and costs for the competing modes, the traditional assumption that existing mode shares will continue in the future is appropriate. If qualitative but not quantitative information is available, it is possible to use that qualitative information to change specific mode shares. Market segments of particular commodities, O-D pairs by shipping distances, and shipment size may be identified and expert opinion used to change the modal share. A typical commodity-based mode split model uses modal shares from the base year commodity data by origin, destination, and commodity group to determine the mode split in the fore- cast year. These base shares are usually not sensitive to factors like travel times, travel costs, safety, and reliability. However, in some instances mode-specific information from the commod- ity data is used to develop freight mode split models. A detailed explanation of these methods is provided in the mode split sec- tion of the O-D factoring method (Section 6.2). The Indiana Commodity Transport Model uses the 1993 CFS data to project observed national modal shares into the future. The mode split model in the Florida model is based on an incremental logit choice model and historical mode split percentages. The base year water and air mode splits for each commodity group are assumed to remain unchanged in the future. The choice model is applied to the splits between truck, intermodal, and carload rail, which pivot about the base year percentages: where Si′ = new share of mode i; Si = original share of mode i; and ΔUi = utility of mode i in the choice set J (j = 1,2,3,. . .,J). The coefficients of the utility function were adopted from a study in New York and calibrated to the TRANSEARCH database for Florida. At the national level, the Vehicle Inventory and Use Survey (VIUS) data set provides a large sample that can be used to determine average payloads by commodity, operating radius, vehicle size, and type of truck usage. This information is ap- plicable to long trips (greater than 200 miles), since these are typically interstate movements. For shorter trips beginning and ending within the state, average payloads should be esti- mated from only those vehicles based in-state. This method has been used widely in many statewide and regionwide freight models. However, there are some exceptions where the freight tonnage is divided into an equivalent number of ′ = ( ) Δ = ∑ S S U S U i i i j I J j exp exp Δ 1 ( ) 13 Mode Split Multimodal Matrix Rail Matrix Truck Matrix Water Matrix Air Matrix 3 4 4 4 4 Figure 4.4. Mode split.

vehicles, with ton-per-vehicle rates determined separately for each commodity group. These rates are based on values (by commodity group) from the Surface Transportation Board Rail Waybill sample and the assumption that each truckload carries 40% of the load carried by a railcar. 4.5 Traffic Assignment As shown in Figure 4.5, the assignment model component uses the table or matrix of freight flows by mode between all zones produced by the mode split model component to fore- cast freight volumes on individual links of the modal net- works. The assignment model component customarily processes each mode separately using a network for that mode with attributes important to freight in order to find the optimum path or sequence of links between all geographic zones. For truck freight flows, the travel times on the highway network may account for the congestion caused by passenger autos and other vehicles. In that case the freight truck trip tables will be assigned together with those auto tables to find the total link travel times and volumes. For the economic activity model class, the link volumes are used to adjust the original economic forecast in an iterative process until an equilibrium is reached. Network assignment models, as used in statewide freight forecasting, apply the modal freight trips to paths identified from the modal network. Essentially three types of assign- ment models are used: rules-based assignment, freight truck only network assignment, and multiclass network assign- ment. Rail networks are typically rules-based assignment models, given the difficulties of including rail business prac- tices in an assignment model. Freight truck only mode and multiclass assignments typically apply only to trucks on high- ways. Rules-based assignment techniques may be developed by the analysts or purchased as part of the existing O-D survey. The distinguishing feature of a rules-based assignment is that the analyst does not have the ability to change the paths to be used in response to changes in performance on the system or the introduction of new facilities. As part of its TRANSEARCH commodity flow database, Reebie Associates provides the option to map truck freight flows on a highway network. This routing is accomplished through the use of special files that contain: • A highway network with unique highway identifiers for each highway segment; • A set of paths between origin and destination zones con- sisting of the highway links used to travel from origins to destinations; and • An O-D table of truck flows by commodity with the iden- tifier of the path used by those flows. The TRANSEARCH Highway Network is available as a Microsoft Access table and an ArcView shapefile. By initiat- ing queries within Access and exporting those results to ArcView, it is possible to develop maps of the flows of some or all commodities on the highway system. In freight truck only assignments, the freight truck trip table is assigned to the highway network using an all-or- nothing assignment process. Since a straight all-or-nothing assignment typically loads too many trips onto the interstate highways, a procedure to adjust the link speeds for noninter- state highway segments is often applied. This serves to draw more trips from the interstate roads to the competing U.S. and state highways that run parallel to them. The unfortunate part of the assignment step is the failure to address the possi- bility of congestion due to the presence of a large number of passenger vehicles sharing the road. The Freight Analysis Framework (FAF) uses a methodol- ogy to estimate trade flows on the nation’s highway infra- structure, seeking to understand the geographic relationships between local flows and overall transportation. Truck assign- ment in the FAF is accomplished using TransCAD’s Stochas- tic User Equilibrium and with other vehicles, such as auto- mobiles, preloaded on the network. FAF is an improvement over the all-or-nothing assignment because it accounts for congestion. 14 5 5 5 5 4 4 4 4 Highway Assignment Rail Assignment Water Assignment Air Assignment Rail Matrix Truck Matrix Water Matrix Air Matrix Rail Volumes Truck Volumes Water Volumes Air Volumes Figure 4.5. Traffic assignment.

Multiclass network assignment of the truck trips can be based on an equilibrium highway assignment, and truck trips are usually assigned together with the passenger vehicle model because congestion has a significant impact on truck travel times. Truck trips may also be assigned separately by vehicle size using the multiclass assignment technique. Many truck models are developed using a conversion of truck vol- umes to passenger car equivalents (PCE) for assignment pur- poses. This factor provides a means of accounting for the fact that larger trucks take up more space on the roads than pas- senger cars, and behave differently during acceleration and braking. This is important to determine the effects on capac- ity and congestion for assignment of both trucks and passen- ger cars. The Transportation Research Board’s Highway Capacity Manual recommends PCE values of 1.5 and 2.0 for single unit trucks with six or more tires and combination units respectively. The truck model developed by the Balti- more Metropolitan Council indicated that the PCE value for heavy truck varies from 2.0 to 4.0. This value depends on roadway grades, acceleration, and braking times. If observed data on passenger car equivalents are collected, then these assumptions by truck type should be modified. 4.6 Economic/Land Use Modeling The economic/land use modeling component shown in Figure 4.6 may be used to prepare the basic socioeconomic forecasts by geographic area used in freight forecasting. If the economic forecasts are prepared independently of the freight transportation forecasts they will serve as inputs to factor facility flow or standalone commodity trip tables or as inputs to the trip generation model component. The forecasts depend on the relative accessibility between geographic zones and the forecast is revised based on the resulting forecast of link volumes. These iterative adjustments may be made as part of a formal model process. Economic/land use modeling components in statewide freight forecasting include modeling techniques known as a spatial input-output (I-O) or econometric models. The land use considerations in these models that consider state and national economic activity are generally far less developed than in metropolitan land use models and typically only fore- cast household and economic activity across county-level zones based on basic supply, demand, and cost relationships for the state and national economy. These models may be used to develop the forecast socioeconomic variables that will be used by the freight model. Econometric models are seldom maintained and operated by state departments of transportation. Most often they are oper- ated by other state agencies, by state universities (such as, Uni- versity of Florida Bureau of Economic and Business Research, University of Kansas Econometric Model) or by private firms (for example, Global Insights as WEFA for the FAF and Ohio Interim Model and as DRI for the Nebraska Model, REMI, Woods & Poole for Indiana). When operated by others, the state departments of transportation may receive and use only economic activity outputs, such as employment by industry and population, to use directly in statewide freight forecasting. Alternatively, they may receive growth rates to apply to existing freight flows, or a complete forecast of future freight flows. When included as a component within the economic activity class of models, the economic/land use model com- ponents may be operated by the state DOT in cooperation with economic development agencies. 15 1 Data Socioeconomic Forecasting Methods Figure 4.6. Economic/land use modeling.