National Academies Press: OpenBook

Statewide Travel Forecasting Models (2006)

Chapter: Appendix D - Annotated Bibliography of Statewide Freight Forecasting

« Previous: Appendix C - Literature on Statewide and Intercity Passenger Travel Forecasting
Page 87
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 87
Page 88
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 88
Page 89
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 89
Page 90
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 90
Page 91
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 91
Page 92
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 92
Page 93
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 93
Page 94
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 94
Page 95
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 95
Page 96
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 96
Page 97
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 97
Page 98
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 98
Page 99
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 99
Page 100
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 100
Page 101
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 101
Page 102
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 102
Page 103
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 103
Page 104
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 104
Page 105
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 105
Page 106
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 106
Page 107
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 107
Page 108
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 108
Page 109
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 109
Page 110
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 110
Page 111
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 111
Page 112
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 112
Page 113
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 113
Page 114
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 114
Page 115
Suggested Citation:"Appendix D - Annotated Bibliography of Statewide Freight Forecasting." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 115

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.

88 This appendix presents material originally developed for NCHRP Project 8-43, “Methods for Forecasting Statewide Freight Movements and Related Performance Measures.” The appendix was written by Alan J. Horowitz, K. Ian Weisser, Cheng Gong, and Joe Blakeman. INTRODUCTION A review of current planning practice indicates that the field of statewide travel forecasting is still in flux; a con- sensus does not exist as to the best way to construct a model for any given set of policy needs or planning requirements. States modeling efforts fall into one of these four categories: 1. No model—Specialized studies work from existing proprietary or public databases or from locally col- lected data. 2. Truck model—Truck models are used to account for the congestion effects of freight on highways or to help determine equivalent single-axle loads (ESALs) for pavement design purposes. 3. Commodity-based four-step model—Commodity- based models follow the same steps as passenger mod- els, except that trip generation is performed for weight of commodities by groups of commodities. 4. Economic activity model—Economic activity mod- els trace the flows of commodities between eco- nomic sectors and between zones. Economic activity models are often implemented within a framework that also forecasts the locations of employers and residences. There is considerable variation in how statewide freight models have been implemented. In addition to those models currently being used by states there is a large variety of models that have been implemented for such purposes as international trade, national trade, energy policy, and corridor studies. Fur- thermore, there are older statewide freight models that have been inactivated, models currently under develop- ment, international freight models, general guidelines as to how statewide freight models may be built, numerous aca- demic studies that attempt to improve freight forecasting methodology, time–series methods of directly forecasting vehicular traffic on facilities, and research and develop- ment intended for passenger forecasting that carries over to freight. OVERVIEWS OF STATEWIDE TRAVEL FORECASTING There have been two notable attempts to define the scope and content of a statewide freight model. The first attempt was NCHRP Report 260: Application of Statewide Freight De- mand Forecasting Techniques, and the second attempt was the Guidebook on Statewide Travel Forecasting. Before dis- cussing these two reports, it is necessary to define “OD table factoring and assignment” as a widely used methodology of statewide freight forecasting. Origin–Destination Table Factoring and Assignment A frequently used method of freight forecasting can be de- scribed as origin–destination (OD) table factoring and as- signment. This method (with some variation) has been used by many states, the I-10 corridor study, and FHWA’s Freight Analysis Framework (FAF). The most prevalent application of this method follows these general steps. 1. Obtain base year OD tables (in tons per year) by com- modity and by mode that matches the desired traffic zone system. Typically, flows between external zones that do not pass though the internal portions of the net- work are excluded. 2. Obtain base year and future year levels of economic activity (by industrial sector) for all zones. 3. Establish a mapping between industrial sectors and commodity categories, such that a percent increase in an industrial sector can be associated with a percent in- crease in a commodity. 4. Determine the percent increase in each commodity’s origins and destinations by applying growth factors obtained in steps 2 and 3. 5. Apply Fratar factoring to each OD table to achieve the percent increases determined in step 4. 6. Determine the number of vehicles necessary to carry each OD flow for one equivalent weekday. 7. Assign each factored vehicle trip table to its respective modal network. This method assumes that the mode split for any given commodity and for any given OD pair is a constant. Any modal shifts that occur in this method are the result of growth (or decline) or spatial shifts in economic activity and the con- sequential effects on commodity production and consump- tion patterns. Shifts owing to changes in costs, supply chain practices, shipping and transfer times, or vehicle technology are not included. APPENDIX D Annotated Bibliography of Statewide Freight Forecasting

89 The method further assumes that the production, con- sumption, and shipping characteristics of commodities re- main unchanged. Such assumptions can be eliminated by careful consideration of changes in (a) shipping density of commodities, particularly the result of packaging materials; (b) worker productivity when economic activity forecasts are given in number of workers in an industry; (c) value per ton when economic activity forecasts are given in monetary units; (d) the routing patterns of the supply chain; and (e) competitiveness of modes or intermodal combinations to carry specific commodities. Those who have tried this method have had to account for important commodity flows that were not included in the original OD tables. In addition, it is necessary to adjust for the number of empty vehicles. NCHRP Report 260: Application of Statewide Freight Demand Forecasting Techniques Memmott, F., NCHRP Report 260: Application of Statewide Freight Demand Forecasting Techniques, Transportation Research Board, National Research Council, Washington, D.C., Sep. 1983, 210 pp. This report was the first major effort to devise a standard method for statewide freight forecasting. The proposed method was based on the generalized procedure of OD table factoring and assignment. A considerable amount of space in the report was devoted to effectively exploiting existing data sources, to forecasting of future consumption of commodi- ties, and to determining the costs of commodity shipment for the purposes of mode split. The report assumes that commodity production is directly related to employment in industries that produce the commod- ity. For estimating consumption, the use of an input/output (I-O) table is recommended. Commodity consumption calcu- lations follow a three-step process: (1) obtain an I-O table, (2) convert dollar amounts to tons and sum the columns of the table to find consumption by industry, and (3) allocate tons to coun- ties [the assumed transportation analysis zone (TAZ) size] according to the employment by consuming industries and population (for final demand) in each county. These steps embody several assumptions, which are explained. The pro- duction and consumption estimates can be applied to an exist- ing commodity-flow matrix or (in the absence of a matrix) incorporated into a gravity model of shipment distribution. Methods of forecasting industrial activity are described. For mode split, the assumption is made that all shipments between a pair of counties of a given commodity are allo- cated to lowest cost mode among those available between the pair. The report goes on to develop a procedure for estimat- ing the cost of shipment by truck, railroad, and barge. No mention was made of air freight or intermodal. All of the cost data are now obsolete, but the terms included in the cost equations are still relevant. Truck cost is composed of insurance, driver wages and benefits, driver expenses, fuel, overhead, licenses and per- mits, ton-mile taxes, federal highway user taxes, tractor cap- ital cost, tractor maintenance, tractor tire cost, trailer capital cost, trailer maintenance, trailer tire cost, stop and delay costs, and terminal cost. The recommended method of rail costing was the Uniform Rail Costing System developed for the Interstate Commerce Commission, which needed a fair method for setting tariffs. Few details are provided on the operation of the computer program, which performs its cost estimates by referencing an extensive database of actual rail costs. The program reports line-haul costs, terminal costs, freight car costs, cost of spe- cialized services, and costs of loss and damage. Barge cost is composed of many components including terminal costs, ownership costs, towing costs, and switching costs. The barge cost module assumes an empty backhaul. Highly detailed information is required about the conditions of the shipment, including the specific origin and destination, tons per barge, towboat horsepower, barge investment, inter- est rates, and user fees. Statistical rate equations to estimate tariffs for both truck (private, truck-load, and less-than-truck-load) and rail (trailer- on-flatcar and carload) are provided. These equations use such independent variables as distance, shipment size, value of the commodity, density, region of the country, rail car ownership, state of matter (liquid, gas, and particulate), and type of terminal at beginning and end of the haul. Shipper costs are added to the modal costs and represent the addi- tional logistics cost borne by the shipper when choosing a specific carrier or mode. These costs include loss and dam- age, pick up and delivery, ordering, warehousing, inventory, and the possibility of running out of stock. Guidebook on Statewide Travel Forecasting Horowitz, A.J., “Freight Forecasting,” Chap. 4, In Guide- book on Statewide Travel Forecasting, Report FHWA-HEP- 99-007, Federal Highway Administration, Washington, D.C., July 1999. Most of this guidebook relates to passenger travel fore- casting, but one chapter deals exclusively with freight fore- casting. This chapter outlines a general method for statewide freight forecasting and draws a distinction between statewide and urban freight models. The chapter is organized accord- ing to the four steps of a standard urban transportation plan- ning model (trip generation, trip distribution, model split, and traffic assignment) plus network development. At each step, the report emphasizes the need to use existing secondary data sources. The general method has 10 steps.

90 1. Obtain modal networks, 2. Develop commodity groups, 3. Relate commodity groups to industrial sectors or eco- nomic indicators, 4. Find base year commodity flows, 5. Forecast growth in industrial sectors, 6. Factor commodity flows, 7. Develop modal costs for commodities, 8. Split commodities to modes, 9. Find daily vehicles from load weights and days of operation, and 10. Assign vehicles to modal networks. A range of options is suggested for many of these steps. Some specific recommendations are as follows: • Spatial unit of analysis—counties are the most conve- nient spatial unit within states. • Networks—network should cover all 48 contiguous states, but focus on the state of interest. Modal networks outside of the state of interest can be adapted from sec- ondary sources. • Selection of modes—modes should be defined consis- tent with the Commodity Flow Survey. • Selection of commodity groups—commodity groups should be developed from Standard Transportation Com- modity Codes (STCCs) or Standard Classification of Transported Goods, disaggregated to the two-digit level. • Trip generation—good production generation relation- ships for commodities can be established by relating industry output to an economic indicator for that indus- try, such as employment. Good consumption generation relationships can be developed by applying the data in an I-O table. • Trip distribution—a gravity model is a good way of representing commodity flows between the production and consumption zones. Such a model can be calibrated to existing data, such as the Commodity Flow Survey. • Mode split—mode split can be handled by a number of techniques, but the complex cost calculations of NCHRP Report 260 should be avoided. Mode split techniques include application of historical fixed shares, aggregate demand formulations, the logit rela- tion, the pivot-point relation, and elasticity methods. • Traffic assignment—all-or-nothing traffic assignment is recommended; however, impedances should be ad- justed to account for biases caused by shippers defining an optimal route differently from the shortest path as in- dicated by traffic speeds. EXTENDED EXAMPLES OF STATEWIDE FREIGHT FORECASTING MODELS: OTHER NOTABLE STATEWIDE FREIGHT MODELS Oklahoma Model TranSystems Corporation, Oklahoma Statewide Intermodal Transportation Plan Freight Report, Oct. 2000. This document is included as Appendix B in the Oklahoma Department of Transportation’s Oklahoma Statewide Intermodal Transporta- tion Plan, Feb. 2001. This model and forecast system were developed in 2000 by TranSystems Corporation. It is a conventional model based on Reebie TRANSEARCH data, but with a major dif- ference in usability. The consultant built a calibrated truck trip model onto 31 corridor segments. The Oklahoma Department of Transportation (ODOT) uses a spreadsheet with the 31 cor- ridors, upon which it can change growth factors or update truck volume. Although the spreadsheet cannot handle major changes in the network or economy, it is a very useful tool for day-to-day forecasting. For clarity, “model” refers to the TranSystems Corpora- tion model and work, whereas “spreadsheet” refers to the corridor truck forecasting system based on the model. Tran- Systems owns the model. ODOT uses only the spreadsheet. The TranSystems model was built to identify the major freight corridors in the state. The model uses three zones within Oklahoma and eight zones outside the state to allocate Reebie data for external–external and external–internal trip tables. The same data identified the internal–internal trips between the three zones within Oklahoma. The study did not attempt to capture county-to-county trips or any scale finer than the three zones. Each commodity is identified by modal split in the data. The network includes only major corridors. The model was calibrated using 1996–1998 ODOT truck volume counts. Additional local detail was introduced by segmenting the network corridors near cities. For example, Reebie data on I-35 show long distance truck movements, which can be calibrated. Then, the heavier urban and suburban truck movements were picked out as part of the calibration process. These urban and suburban areas were placed in different corridor segments, so that an urban- rural-urban corridor would be three segments; high-low-high volume. In this way, some short trips in the corridor can be indirectly modeled using the intercity data. Rail, air, and water trips are included in collected data, but are filtered out in the modeling process. ODOT keeps statis- tics on all modes, but models only truck movement. I-66 Southern Kentucky Corridor Wilbur Smith Associates, “Kentucky Statewide Traffic Model Final Calibration Report,” Apr. 1997. Wilbur Smith Associates, “Kentucky Statewide Traffic Model Update,” Jan. 2001.

91 Model Structure This model was developed in 1997 by Wilbur Smith Associ- ates. The network and base data were updated in 2001 by Wilbur Smith, without changing the model methodology. The model has 1,530 traffic analysis zones; about half in Kentucky and half in surrounding states. TAZs are based on groups of census tracts. The model includes TAZs up to 3-h drive time outside Kentucky, including St. Louis, In- dianapolis, Cincinnati, Columbus (Ohio), Nashville, and Memphis. The Kentucky statewide model is truck only, and does not include rail, marine, or air freight. It uses Reebie Associates data and cordon count data to determine truck trip generation. Future forecasts also use Fratar factors. All truck trips ends in each county are assigned to a sin- gle TAZ. This is different from automobiles, which can have more than one TAZ per county. Internal Trips Reebie Associates data are disaggregated from 56 zones across North America, plus 28 zones in Kentucky, to 469 Kentucky model TAZs (maximum of one TAZ per county). Disaggregation is based on population and em- ployment. Assumptions: equal truck trips daily (including weekends and holidays), and uniform weights of 16.8 tons per truck, regardless of commodity. Reebie data assumes that inbound and outbound trips and tonnage for each zone are not equal, but that total inbound and outbound sums of trips and tonnage for the entire Kentucky Model area are equal. The resulting inbound and outbound trips for each county are not used directly, but become the baseline for the internal trip gravity model. This gravity model determines the internal truck trip table. External Trips External trips are based on cordon counts and surveys con- ducted in Ohio (1996) and on traffic counts. The cordon counts and surveys include autos and trucks. The cordon surveys show the external–external trips and provide the basis of distribution for the external–internal trips. The dis- tribution of all external–internal trips is assumed to match the survey results. All external trips are assumed to be symmetrical, with one outbound trip matched by one in- bound trip. Finally, the volume of external trips comes from existing traffic count data at each “entry station,” where the trip enters and leaves the model. This volume and distribution becomes the basis for the external– external and external–internal truck trip tables. Trip Table Calibration This model does not have a method to calibrate truck trip tables. Trip Assignment Trip assignment is based on user assumptions and desired reports. Wilbur Smith Associates describes trip assignment as “. . . the least complex part of the [model].” Growth Factors Local truck trip growth is based on projected population and employment growth in that county. A Fratar model is used to apply the factors. The Network The network within Kentucky was developed on MINUTP. The network outside Kentucky was developed from National Highway Planning Network Version 2.0. The entire model network was migrated to TransCAD in 2001. Updating the Model The model can be updated with new Reebie data, new exter- nal distribution survey or truck volume data, and new Woods & Poole population and employment data. Vermont Cambridge Systematics, Inc., “Vermont Statewide Freight Study,” Final Report, prepared for the Vermont Department of Transportation, Montpelier, Mar. 2001. Cambridge Systematics developed a complete freight forecasting model as part of the Statewide Freight Study. This model follows a variation of the classic four-step model. OD data included Reebie TRANSEARCH data, roadside surveys, motor carrier surveys, and interviews with key ship- pers. Link data included traffic recorder and weigh-in-motion detector, plus data from previous local and corridor studies. Future commodity-flow patterns were developed by Standard and Poor’s DRI for years 2005, 2010, and 2020. The network was created from 14 in-state zones and 16 out-of-state zones. The network links and nodes are not presented in the document. Annual commodity flows were converted into truck move- ments using data from the 1997 Vehicle Inventory and Use Survey from the U.S. Census Bureau. The county-to-county truck trip tables were built from DRI forecasts, Reebie data

92 and survey data. The truck trip tables were then converted to passenger car equivalents for assignment to the highway network. After the highway assignment and a complete run of the model, mode split between truck and rail is determined using a sensitivity analysis based on the roadside surveys. Mode split is tailored for each region (method not explained). The resulting changes to OD tables can be compared with the truck-only model. Appendixes detail Reebie data, surveys and interview formats. Kansas Russel, E., L. Sorenson, and R. Miller, “Microcomputer Transportation Planning Models Used to Develop Key Highway Commodity Flows and to Estimate ESAL Val- ues,” unpublished, prepared for the Midwest Transportation Center at Iowa State University and the Kansas DOT. This network uses General Network Editor and Quick Re- sponse System II (QRS II) to model the flow of five agricul- tural commodities in Kansas. The network uses 202 zones and 2,200 links. The purpose of the model is to determine truck volume and axle weights (ESALs) for improved pave- ment design. Data include existing K-Trans surveys and commodity data provided by Kansas State University. The model has not been validated and is not being updated. The study team of- fered recommendations on improved commodity weights, link speeds, and other network and model changes. Nebraska Jones, E. and A. Sharma, “Development of Statewide Freight Forecasting Model for Nebraska” (CD-ROM), Transporta- tion Research Board, National Research Council, Washing- ton, D.C., 2003. The authors use standard four-step modeling techniques for a statewide model based on the Wisconsin model, but introduce a separate method for agricultural commodities based loosely on the Kansas model. Trip productions used normal data sources, such as the 1993 Commodity Flow Sur- vey. IMPLAN software provided the I-O coefficients used to derive trip attractions. Agricultural shipments were modeled separately from other commodities to enable analysis of intermodal grain transportation. Agricultural surveys and data sources were used to accurately determine commodity productions for each zone. Production, elevator locations, and capacity and rail service determined mode split and trips for each agricultural commodity. Model details and algo- rithms are not provided. Virginia Brogan, J., S. Brich, and M. Demetsky, “Identification and Forecasting of Key Commodities for Virginia,” Transporta- tion Research Record 1790, Transportation Research Board, National Research Council, Washington, D.C., 2002, pp. 73–79. The paper shows a model-building method based on only major commodity flows. It also includes lessons learned from the first step of Virginia’s freight planning methodol- ogy. The freight planning methodology includes several model elements, not a complete forecasting model. Virginia’s six-step method of freight planning is: 1. Inventory the system—the lessons learned are from this step. 2. Identify the problem. 3. Establish performance measures. 4. Collect data for specific problems. 5. Develop and evaluate improvement alternatives. 6. Select and implement improvements Rather than construct a model of all freight flows, the Vir- ginia DOT purchased Reebie TRANSEARCH data and eval- uated only the 15 top commodities based on weight or value. Once these commodities were identified, each was assigned to a set of OD matrices. The matrices are input to IMPLAN soft- ware with an integral employment database, creating relation- ships between commodity flow, employment, and dollar value. Comparing employment, population, and other factors with commodity production and consumption, the authors used a set of regression techniques to determine production factors and consumption factors for each key commodity. This way, changes in employment or related industries can be con- verted into changes in tons of commodity flow. The commod- ity flows are assigned to a statewide network, beyond the scope of the paper. No single regression technique worked well for identifying generation of consumption factors of all types of generators or consumers on the network. Port facilities behaved very differ- ently from other types of facilities. Variables including total employment and transportation employment were important factors for some commodities. Freight consumption models were more accurate than freight production models. Factors behind freight mode choice were not clear. Louisiana Apffel, C., J. Jayawardana, A. Ashar, K. Horn, R. McLaugh- lin, and A. Hochstein, “Freight Components in Louisiana’s Statewide Intermodal Transportation Plan,” Transportation Research Record 1552, Transportation Research Board, Na- tional Research Council, Washington, D.C., 1996, pp. 32–41.

93 The planning procedures for the freight components of Louisiana’s Statewide Intermodal Transportation Plan for water, rail, and intermodal components are presented. The planning included U.S.DOT’s four Cs (connection, choice, coordination, and cooperation), as well as reflecting actual freight movements. The Louisiana Department of Transportation and Devel- opment set a 25-year planning horizon for the study, which included input from both freight users and providers. Low- cost and high-capital-cost improvements were considered for addressing capacity issues. These improvements were then evaluated by a sample of potential users and compared with goals. A roster of statewide freight users was assembled to submit a draft of challenges to the staff for defining statewide inter- modal needs. This statement was revised as input from con- current technical analysis was presented. In addition, industry executives were interviewed to provide diverse perspectives. For flow analysis, four types of trips were included, internal– internal, internal–external, external–internal, and external– external. Historic data were gathered from the U.S. Army Corps of Engineers (waterborne commerce statistics), Interstate Commerce Commission (rail waybill information), and Reebie Associates (TRANSEARCH database). The raw data were aggregated into business economic areas (BEAs) within the state and super BEAs for states outside Louisiana and interna- tional markets. Commodities were broken into 11 categories based on their nature, transport, handling requirements, etc. All com- modity movements were analyzed in terms of modal share, origins and destinations, and domestic or foreign trade flows so that factors affecting future growth could be identified and assessed. Relational database systems were built to extract and aggregate flow measures by commodity groupings, by mode, and by BEA or super BEA origins and destinations. Demand projections focused on three growth scenarios, high, medium, and low. High and low were generated from forecasts for 1990–2000 made by federal agencies and in- dustry groups. The growth rates were adjusted downward for all commodities beyond 2000 to incorporate long-term un- certainty. The study included three 10-year periods: 1990–2000, 2001–2010, and 2011–2020. Evaluations were made for existing and future production to determine future market developments and augment earlier information provided by static quantitative growth forecasts. The analysis included specific industrial sectors, productivity trends, and the competitive position of transportation providers in the state. Freight network analysis focused on transshipment fa- cilities and intermodal connections. The methodology used included detailed capacity measures of the different trans- fer and storage aspects of intermodal terminals for rail–highway and marine facilities, augmented by analysis of the performance of terminal access routes that provide intermediate linkages to corridor and line haul routes. Terminals used for capacity assessment included five generic types for water and rail-based freight. The capacity was determined using stock and flow analysis of terminal op- erations. The study determined that capacity was the product of two factors—effective transfer rate (tons/day) and effec- tive working time (days/year). Terminal access was also studied by use of a detailed in- ventory and assessment of intermodal terminal accessibility. An inventory of the characteristics of local access roads and rail spurs was made for public marine and rail–highway ter- minals in the state. Questionnaires were distributed to every operator of these terminals in the state and were supple- mented by field surveys as necessary to document the phys- ical, institutional, and operating aspects of terminal access. Iowa Iowa State University, “Developer’s Guide for the Statewide Freight Transportation Model,” Iowa State University Center for Transportation Research and Education, undated [Online]. Available: http://www.ctre.iastate.edu/Research/ statmod/dev_guid.pdf. This TRANPLAN-based model is not a forecasting tool, but is instead used for policy analysis. As with other similar models, it uses Reebie TRANSEARCH commodity data, or- ganized by BEA zones, connected by TRANPLAN networks for multiple modes, and organized by standard industrial commodity codes. BEA zone data are disaggregated to county level using NCHRP Report 260 techniques. The model uses a simple gravity model, with different fric- tion factors for food and machinery. Only internal–internal and internal–external trips are modeled. External–external trips are assumed to be beyond the policy applications of the model. The model uses separate network layers for road and rail modes. The road network is typical of other plans. The rail network is subdivided by carrier, and interchanges between carriers are limited to actual interchange points defined by in- put from the rail carriers. Because most software is designed for roads, the rail network noninterchange nodes were as- signed turn penalties. Impedance for road, rail, and inter- modal movements are based on cost only, not time. The model can be used for evaluation of changes in trans- port cost, production or consumption, and infrastructure (network).

94 SIMPLER METHODS Simpler methods are intended for rapid application of exist- ing data to determine one or a few forecasted items. Usually intended for short-term forecasts, many assumptions are needed to make them work and their range of applicability is limited. Simpler Methods from the Guidebook on Statewide Travel Forecasting Horowitz, A.J., Guidebook on Statewide Travel Forecasting, Report FHWA-HEP-99-007, Federal Highway Administra- tion, Washington, D.C., July 1999. Time–Series Methods—The Guidebook on Statewide Travel Forecasting discusses time–series methods for direct forecasts of vehicular volumes on highways and for fore- casting the inputs to four-step models. Major emphasis is on ARIMA (autoregressive integrated moving average) models and on growth factor methods. Examples are primarily for passenger car forecasting; however, the methods are equally applicable to truck forecasting. I-40 Truck Model—The Guidebook describes a linear regression model to forecast truck volumes on I-40 in New Mexico. Commercial truck traffic was found to be a linear function of the year, the U.S. disposable income, U.S. gasoline costs, and the New Mexico cost of residential construction. Simpler Methods from the Quick Response Freight Manual Cambridge Systematics, et al., Quick Response Freight Man- ual, Report DOT-T-97-10, Travel Model Improvement Pro- gram, Federal Highway Administration, Washington, D.C., Sep. 1996. The Quick Response Freight Manual (QRFM) describes two methods of applying growth factors to traffic volumes that are applicable to rural highways as well as urban high- ways. The first method involves estimating a growth factor from current and past truck count data and applying the re- sulting factor to future years using a conventional compound interest formula. The second method determines several growth factors, one each for many economic indicator vari- ables, usually employment in local industrial sectors. The future growth in economic indicator variables, as calculated by a compound interest formula, is used to forecast growth in commodity groups, which is then used to forecast the growth in trucks carrying each group of commodities. Nec- essary assumptions about the economy and freight charac- teristics are discussed. A similar concept is described in NCHRP Report 388. Truck Model from the Quick Response Freight Manual Although the QRFM was intended for urban forecasts, it out- lines a process that might also be used to create a statewide truck model. The QRFM follows a three-step process of trip generation combined with vehicle split, trip distribution, and traffic assignment. The most interesting aspect of the QRFM is the generation of truck origins and destinations (not pro- ductions and attractions) for each zone by three categories of commercial vehicles: four-tire trucks, other single-unit trucks, and combination trucks. Origins and destinations are linear functions of employment in industrial sectors and numbers of households. Elasticity Methods from NCHRP Report 388 Elasticity and cross-elasticity methods are suggested in NCHRP Report 388 in the appendix, “Rail/Truck Modal Di- version.” Tables of cross-elasticities are given between rail and truck by commodity group as derived from a proprietary model, the Intermodal Competition Model developed by the Association of American Railroads. A cross-elasticity can be interpreted as the percentage change in one mode’s share given a one percent change in an attribute of another mode. For example, it might be possible to estimate a change in the rail share of carrying primary metals from the change in cost of carrying primary metals by truck. Pivot Mode Share Method from the Guidebook on Statewide Travel Forecasting The pivot formulation of a mode split model as found in the Guidebook on Statewide Travel Forecasting was applied in the Florida model discussed earlier. This method can be used as a stand-alone technique to estimate mode shares for local- ized generators. A pivot formulation is able to forecast new mode shares from knowledge of existing mode shares and the change in a single variable in the “utility” of transporting a unit amount of a commodity by a particular mode. The Guidebook recommended using cost as the single variable. Forecasting Based on Cost Data Memmott, F.W. and R.H. Boekenkroeger, “Practical Method- ology for Freight Forecasting,” Transportation Research Record 889, Transportation Research Board, National Re- search Council, Washington, D.C., 1982, pp. 1–7. The freight-demand forecasting technique discussed in this paper is a very simple and straightforward methodology. Compared to a formal mathematics model, the technique de- scribed in this paper is really a process for systematically making a large number of revenue and/or cost calculations. The structure of the model follows these steps: (1) prepare model inputs; (2) compute base case transport costs and

95 revenues; (3) develop alternative futures, scenarios, and con- ditions; (4) compute alternative transport costs and revenues; (5) summarize computed information and print reports; and (6) conduct highway impact analysis. Computations for dif- ferent states involve either hand or computer calculations. By adding some other components, the structure can be modi- fied for use in different transport models. The most impor- tant inputs are origins and destinations of the movements or flows and the unit costs and revenues. The paper gives some formats for unit costs and revenues. Two examples are de- scribed: one is for a Montana study and the other is for a U.S. Army Corps of Engineers study. Control information, com- modity flows, revenues or charges, costs, unit distances, and vehicle equivalents are needed for study. Highway impact analysis is also discussed in this paper. Application of Regression and Elasticity Techniques Morton, A.L., “A Statistical Sketch of Intercity Freight Demand,” Highway Research Record 296, Highway Research Board, National Research Council, Washington, D.C., 1969, pp. 47–65. Time–series regression analysis is used to estimate de- mand for truck and rail. The first part of the paper describes data, which were organized into five commodity groups: agriculture, animals and products, products of forests, prod- ucts of mines, and manufactures and miscellaneous. The rail and truck price index and truck rate series were needed to estimate the price and the cross-price elasticity of demand. The gross national product (GNP) was used to estimate the income elasticity of demand. A logarithmic form of a re- gression equation was selected after three demand equations were estimated, both logarithmically and untransformed. It was assumed that one year is long enough to estimate the traffic volumes by using the previous year’s prices. In total, 12 markets were studied; two sets of equations and 324 coefficients were estimated. For rail the growth in GNP gen- erated new traffic at the level of three-fifths of the rate of economy expansion. Economic growth generated new traffic for truck at double the rate of economy expansion, and truck traffic was more influenced by price. OTHER RELEVANT NCHRP STUDIES NCHRP Report 177 Roger Creighton Associates and R.L. Banks and Associates, NCHRP Report 177: Freight Data Requirements for Statewide Transportation Systems Planning: Research Report, Trans- portation Research Board, National Research Council, Wash- ington, D.C., 1977. This report reviewed data needs and data availability for statewide freight planning during a period of time when freight planning was in its infancy. Because many of the data sets no longer exist or have changed in character, the specific recommendations about them are no longer relevant. In addition, methods for freight planning have also changed substantially. Still of current relevance is a matrix for each mode (rail, truck, ports, inland waterways, pipelines, and air cargo) that relates planning issues to data needs. The report identifies and describes 64 planning issues that could be better addressed by analysis of freight data. NCHRP Report 178 Roger Creighton Associates and R.L. Banks and Associates, NCHRP Report 178: Freight Data Requirements for Statewide Transportation Systems Planning: User’s Manual, Transportation Research Board, National Research Council, Washington, D.C., 1977. This report describes ways to implement the findings of NCHRP Report 177, a companion report to this one. Most of the database descriptions in this report are now obsolete; however, the authors have provided general guidance on how to use freight data in transportation planning that is still quite useful. The report outlines a procedure for identifying freight data requirements consisting of these steps: 1. Identify Freight Issues and Problems, 2. Arrange Issues and Problems in Priority Order, 3. Establish Planning Program for Freight Transportation, 4. Determine Planning Methods, and 5. Determine Data Needs and Available Resources. The report further discusses the advantages of assem- bling data from secondary sources and ways of obtaining primary source data. Primary source data includes traffic flow data, carrier data, shipper and consignee attributes, physical and operational data, and direct and indirect impacts. A lengthy appendix provides guidance on how to organize and conduct a shipper survey, one of the possible primary data sources. NCHRP Report 388 Cambridge Systematics, Inc., et al., NCHRP Report 388: A Guidebook for Forecasting Freight Transportation Demand, Transportation Research Board, National Research Council, Washington, D.C., 1997. NCHRP Report 388 is intended as a guidebook to help planners perform freight planning and forecasting. It gathers reference information about freight transportation planning processes, techniques, tools, data, and applications. The first chapter of the report describes its purpose, the characteristics of the freight demand, the current study, and some related

96 research. The second chapter describes factors that influence freight demand. The third and fourth chapters discuss de- mand forecasting for both existing and new facilities. New facility options include new highways for serving rail yards, new rail facilities for current railroads, new rail facilities for competing railroads, and new U.S. or foreign port terminals. The last chapter describes policy analysis. The report’s appendixes contain a wealth of useful infor- mation; factors influencing freight demand, reviews of freight-demand forecasting studies, freight activity data sources, freight transportation survey procedures and meth- ods, statistical forecasting techniques, estimating transport costs, rail and truck modal diversion, three modal-diversion models, case studies, and the information needs perceived by public agencies. Appendix B is an annotated bibliography of several of the more important freight-demand studies. • Cambridge Systematics, Inc., Alternative Planning Ap- proaches: Structural and Direct, NCHRP Project 20-17, Statewide Freight Demand Forecasting, May 1980. • Memmott, F. and Roger Creighton Associates, NCHRP Report 260: Application of Freight Demand Forecasting Techniques, Transportation Research Board, National Research Council, Washington, D.C., 1983. • Memmott, F.W. and R.H. Boekroeger, “Practical Methodology for Freight Forecasting,” Transportation Research Record 889, Transportation Research Board, National Research Council, Washington, D.C., 1982, pp. 1–7. • Kim, T.J. and J.J. Hinkle, “Model for Statewide Freight Transportation Planning,” Transportation Research Record 889, Transportation Research Board, National Research Council, Washington, D.C., 1982, pp. 15–19. • Middendorf, D.P., M. Jelavich, and R.H. Ellis, “Devel- opment and Application of Statewide, Multimodal Freight Forecasting Procedures for Florida,” Trans- portation Research Record 889, Transportation Re- search Board, National Research Council, Washington, D.C., 1982, pp. 7–14. • Hu, P., T. Wright, S. Miaou, D. Beal, and S. Davis, Es- timating Commercial Truck VMT of Interstate Motor Carriers: Data Evaluation, Oak Ridge National Labo- ratory Report, Oak Ridge, Tenn., Nov. 1989, 176 pp. • Friedlaender, A.F. and R.H. Spady, “A Derived Demand Function for Freight Transportation,” Review of Eco- nomics and Statistics, Vol. 62, No. 3, 1980, pp. 432–441. • Lawrence, M.B. and R.G. Sharp, “Freight Transporta- tion Productivity in the 1980s: A Retrospective,” Jour- nal of the Transportation Research Forum, Vol. 32, No. 1, 1991, pp. 158–171. • Winston, C., “The Demand for Freight Transportation: Models and Applications,” Transportation Research, Vol. 17 (A), 1983, pp. 419–427. • Crainic, T., M. Florian, and J.-E. Leal, “A Model for the Strategic Planning of National Freight Transportation by Rail,” Transportation Science, Vol. 24, No. 1, 1990, pp. 1–24. • Stevens, B., Basic Regional Input-Output for Trans- portation Impact Analysis, NCHRP Project 8-15A, Regional Science Research Institute, Philadelphia, Pa., July 1982. • Eusebio, V. and S. Rindom, Grain Transportation Ser- vice Demand Projections for Kansas: 1995 and Beyond, Kansas Department of Transportation, Topeka, July 1990. Some of these reports and articles are also reviewed here. Appendix C contains detailed descriptions of three dozen data sources related to freight transport activity and demand. These include: • 1993 Commodity Flow Survey; • TRANSEARCH; • Freight Transportation and Logistics Service; • U.S. Exports by State of Origin of Movement; • Directory of U.S. Importers and Exporters; • National Transportation Statistics, Annual Report; • Freight Commodity Statistics; • North American Trucking Survey; • LTL Commodity and Market Flow Database; • Nationwide Truck Activity and Commodity Survey; • Ship Movement Database; • Truck Inventory and Use Survey; • State Estimate of Truck Traffic; • Quarterly Coal Report; • Natural Gas Annual; • Surface Transborder Trade-Flow Data; and • Port Import and Export Reporting Service. NCHRP Synthesis 298 Fischer, M.J. and M. Han, NCHRP Synthesis of Highway Practice 298: Truck Trip Generation Data, Transportation Research Board, National Research Council, Washington, D.C., 2001, 81 pp. This report is essential. It describes vital data sources, key considerations for forecasting, many best-practices tech- niques, and many common mistakes made in planning and modeling. Chapter two discusses how truck productions and attrac- tions differ from automobile trips and activities. It also in- cludes a discussion of data collection techniques. Chapter three is an annotated bibliography of data sources, organized by topic. Topics include compendia of trip generation data, engineering studies, special generator

97 studies, ports and intermodal data sources, vehicle-based travel demand models, commodity-based travel demand models, and other critical data resources. Chapter four describes the current (2000) state of the art. A key lack of uniformity in statewide vehicle-based forecasting is the discrepancy between linked (truck with multiple stops) and garage-based (truck with single destination) truck trips. Metropolitan and statewide models treat each differently, making comparisons or pattern identification difficult. Vehicle-based models as a function of employment are popu- lar within metropolitan models. Commodity-based models are more popular at the statewide level and have different prob- lems. Errors in commodity payload factors and other assump- tions are the most common. The report does not recommend methods to standardize or share data among different models (and different orga- nizations), and calls for additional research mostly on im- proved data collection. There is no significant discussion of alternative or lower cost data, new mathematical or computer tools to improve the process or changes, or alternatives to the basic process of trip generation–trip distribution–traffic assignment. The report does not address intermodal activities beyond bibliographic refer- ences to special generator models. ACTIVE NATIONAL OR INTERNATIONAL MODELS AND TOOLS Freight Analysis Framework “The Development of the Freight Analysis Framework Data- base and Forecast,” no date (around 2003), Booz, Allen and Hamilton [Online]. Available: http://www.ops.fhwa.dot.gov/ freight/lambert_files/CombinedFinalMethodologyPiece-2.doc. Fekpe, E., M. Alam, T. Foody, and D. Gopalakrishna, “Freight Analysis Framework Highway Capacity Analysis, Draft Methodology Report,” Battelle, Apr. 18, 2002 [Online]. Available: http://www.ops.fhwa.dot.gov/freight/ lambert_files/Capacity-Method-report-revised.doc. The FAF is a freight forecasting model, developed by FHWA, which covers the contiguous 48 states and the Dis- trict of Columbia. FAF employs the general methodology of OD table factoring and assignment to perform forecasts to 2010 and 2020 from base year OD tables of 1998. GBFM “Great Britain Freight Model” [Online]. Available: http://www.mdst.co.uk/MDSTBody-gbfm.htm. GBFM is an offshoot of STEMM (Strategic European Multimodal Modelling) and contains specific improvements for applications in Great Britain. NEAC The NEAC-Model, “The Solution to Western and Central Eu- ropean Transport Information Problems! Base Year 1997— Forecast 2020” [Online]. Available: http://www.nea.nl/dutch/ publicaties/Brochures/NEAC-folder.pdf. NEAC is a decision support system covering all of Europe that provides the link between traffic and economic develop- ment in and between regions. The advantage of NEAC’s database is that it can determine the exact origin and desti- nation of commodities in region-to-region transport as well as the organization of transport (direct or with transship- ment). Information on the route of shipment can be provided. The NEAC transport chain database can help analysis of transport flows and forecasting based on economical rela- tions. The concept of the transportation chain is described as follows: It is a sequence of transport modes used to carry a certain good from its first origin to its final destination. Along the chain, one or more transshipments take place. NEAC has been applied in some European regions on a range of topics including transport flow analysis, corridor analysis, infrastructure analysis, market potential analysis, and policy impact analysis. EUFRAT PRODEC Resources, “The European Freight Assessment Model” [Online]. Available: http://www.prodec.dk/resources/ eufrat/eufrat.htm. EUFRAT is a multimodal freight assessment model. The network includes all major road, rail, inland waterway, and sea connections, which covers all of the European Union, the European Free Trade Association countries, and all of East- ern Europe, including Russia and Ukraine. It has been ap- plied with reported good results. It uses the common Euro- pean standard for regional statistics (NUTS2, sometimes NUTS3) and the freight volumes are based on the OECD SITC Rev. 2 commodity classes. SAMGODS “National Freight Model System for Sweden” [Online]. Available: http://www.rand.org/publications/MR/MR1663/ MR1663.pdf. SAMGODS is a freight model for Sweden that is cur- rently under development. GTAP “Global Trade Analysis Project (GTAP),” Purdue University, West Lafayette, Ind. [Online]. Available: http://www.gtap. agecon.purdue.edu.

98 Hertel, T. and T. Marinos, “Structure of GTAP,” Draft of Chapter 2, In Global Trade Analysis: Modeling and Applications, Cambridge University Press, 1997 [Online]. Available: http://www.gtap.agecon.purdue.edu/resources/ download/86.pdf. The Global Trade Analysis Project (GTAP) is a complex set of databases and sophisticated economic modeling tools developed by an international consortium of universities, in- stitutions, and government departments in the developed world. The consortium began work in 1993, and research and development work is ongoing. The international consortium approach to developing the model and framework has led to wide use of GTAP economic models for policymaking in the World Trade Organization (WTO), World Bank, and several international conferences. The model uses 57 commodity sectors, some of which do not need freight transport (banking, electricity), but are still considered commodities for purposes of the economic model. Charts show the mapping of these sectors to U.S. STCC groups. Only three types of industrial sectors are iden- tified: agriculture and food, energy, and goods and services. The Earth is divided into 66 regions for determining trade between them. The main uses are to model the effects of economic growth, trade policy changes, and impacts of changes in re- sources, technology, and the environment. The model structure uses generally accepted market-based economic principles, in most cases, to determine number value (as opposed to real quantity or dollar) relationships between producers, consumers, and governments. The set of interlocking relationships causes value to flow and the econ- omy to change or grow, as it would in a real economy. The dynamic flow of value is monitored and held in equilibrium by a layer of accounting within the model. For example, the change in price of commodity X is determined not just by sup- ply and demand, but also by weighted averages of the costs of related commodities and possible substitutable commodities. Tax structures and import and export taxes are included. Because the model uses market-based economic behavior predictions, government fiscal and revenue policy is highly discretionary. Government is also immune to many of the ac- counting checks. Balanced budgets are neither assumed nor required in the GTAP universe. Similarly, the value flows and accounting levels are assumed to be transparent to macroeconomic policies, monetary policy changes, and other non-market-driven events. The feedback from market to pol- icy is political, not economic, and no adequate modeling mechanism exists. Firms purchase land, labor, capital, intermediate inputs, and knowledge (technology) to produce their outputs. Each input varies based on the firm’s location in the global (or re- gional) economy, the skills base, and the readiness of substi- tutes. For example, firms in developed nations tend to use capital-intensive, technologically sophisticated production that minimizes land, labor, and intermediate input costs. De- termining rates of substitution of resources is a major part of modeling firm behavior; for example, purchasing more effi- cient equipment when fuel prices rise. Another example is that intermediate inputs may be locally produced or pur- chased internationally. Finally, each sector of the economy (agriculture, machine manufacture, and banking) has its own special resource needs. Households input government services, income, and goods, and output taxes, purchasing, and savings. Purchase rates by industrial sector are based on birth rates and other criteria. Changes in purchasing are based on elasticities of demand (goods prices), taxes, population change, and other factors. The GTAP world also includes mechanisms for infla- tion and the distribution of multiregional (or transnational) investments. Regions exchange value using the model’s Global Bank and Global Trade mechanisms. Both are monitored only in terms of value. Global Trade does include commodities, but by value instead of tonnage. For use in statewide freight forecasting, GTAP has limited usefulness. The model can provide predicted rates of economic growth within the given regulatory and tariff framework, but converting such models into economic forecasts is already done by the U.S. Department of Commerce and other organizations. Specifically for states with major international port or border activities, GTAP can provide useful commodity value forecasts that can be converted into tonnages. GTAP use as a long-term predictive model is unknown. Its methodology is complicated, but not controversial. It has not seen major use as a predictive model, and has been in use for only 10 years. STEMM “STEMM Final Summary Report” [Online]. Available: http://www.cordis.lu/transport/src/stemmrep.htm, last up- dated April 1999. STEMM (Strategic European Multimodal Modelling) has a multimodal freight model that has flow attributes, includ- ing disaggregation levels and mode and route choice algo- rithms. For each mode and route a generalized cost is calcu- lated by adding financial cost and various qualities of service penalties. Only alternatives within a certain percentage of the lowest generalized costs are considered and if they are the same, then traffic will be split between them.

99 The MDST model was designed according to require- ments of the corridor studies (Cross-Channel and Trans- Alpine). In both cases, traffic is predicted on a limited set of international links where modal change is either essential or at least viable. The STAN (Strategic Transport Analysis) model was used for Nordic case studies. Attribute structure comes from the STEMM Ideal Model Shell and mode and route choices are made using the STAN algorithm. For the Scan Link Cor- ridor case study, the network included subnetworks for nine modes: road, rail, fast rail, truck ferry, car and truck ferry, rail ferry, sea bulk, general sea, and inland waterways. During the analysis, new features were added to STAN, including path analysis. Both MDST and STAN produced plausible results for their respective studies. SOFTWARE PRODUCTS SPECIFICALLY FOR FREIGHT FORECASTING STAN Crainic, T.G., M. Florian, J. Guelat, and H. Spiess, “Strategic Planning of Freight Transportation: STAN, An Interactive- Graphic System,” Transportation Research Record 1283, Transportation Research Board, National Research Council, Washington, D.C., 1990, pp. 97–124. This study details the process used to develop a freight modeling program. The program was coded in ANSI FORTRAN 77 and was developed to be user friendly by standards at the time. It includes modules for input, mod- ification, display, and output for multicommodity net- works. It also includes a network editor, matrix editor, and function editor. The networks are described by modes, products, vehicles, base network, and transfers. The matrices are used to display OD information similar to EMME/2. The functions are used on links and transfers as unit cost functions, and there are a maximum of three for each product in the scenario. STAN uses a general assignment procedure that is a mul- timode multiproduct method that minimizes the total cost of shipping for products considered, from origins to destina- tions, and by means of permitted nodes. Penalties must be added as functions because capacity is not included in the software. This model does not identify shippers and carriers explic- itly, but is used for scenario comparisons when major in- vestments are considered. The shippers’ behavior is therefore only defined in the OD matrix. STAN also does not include an algorithm for simultaneous determination of flows and demand matrices. Crainic, T., M. Florian, and J. Leal, “A Model for the Strategic Planning of National Freight Transportation by Rail,” Transportation Science, Vol. 24, No. 1, pp. 1–24. The authors use a STAN-based network to model the ef- fects of different infrastructure investment scenarios among several modes. The model is trip-based and most detailed in rail operations. The rail model algorithms are included. The entire model is not presented. There is no discussion of fore- casting beyond the effect of differing traffic assignments and mode splits. Guelat, J., M. Florian, and T. Crainic, “A Multimode Mul- tiproduct Network Assignment Model for Strategic Planning of Freight Flows,” Transportation Science, Vol. 24, No. 1, pp. 25–39. This paper describes an economic model, based on cost minimization, for mode split and network assignment. The model is built into the STAN network assignment applica- tion. The network assignment algorithm is presented. There is no discussion of forecasting beyond the effect of differing traffic assignments and mode splits. Guelat, J., M. Florian, and T.G. Crainic, “A Multimodal Network Assignment Model for Strategic Planning of Freight Flows,” Transportation Science, Vol. 24, No. 1, pp. 25–39. The class of models described is network models that pre- dict multicommodity flows over a multimodal network with detail appropriate for large geographical regions or nations. The model described does not consider shippers and carriers as distinct actors, but aggregates all data to origins and des- tinations that are large areas. The model uses a network of links, nodes, and modes, and links are defined as triplets. Each link contains an origin node, destination node, and mode. Transfers are represented by arcs connecting nodes on different modes. The model bases freight flows on least total cost. The model considers a set of nodes, arcs, modes, and transfers. Each arc and transfer is associated with a cost function that relates to the volume of goods on the arc or other arcs in the network. The cost functions that are used to model delays and costs on the links and transfers of a freight network are link sepa- rable except for transportation services that share the same facilities. The code that implements the solution algorithm for the model includes partial derivatives for the computation of marginal costs that is carried out by a precise numerical approximation procedure. The solution algorithm includes the Frank and Wolfe lin- ear approximation method. The shortest path algorithm

100 allows transfers that are similar to turn penalties in urban networks and is an adaptation of Dijkstra’s label setting algorithm. The model is imbedded in the STAN interactive graphic system and has been used successfully in practice in south- eastern Brazil. Crainic, T.G., M. Florian, and J. Leal, “A Model for the Strategic Planning of National Freight Transportation by Rail,” Transportation Science, Vol. 24, No. 1, pp. 25–39. The goal of the article is to achieve an aggregate, strate- gic modeling framework of freight transportation by rail that may adequately reflect the economic and spatial–temporal relations typical to the rail mode. A brief overview of STAN is presented. The network op- timization model that is used to simulate network flows in STAN is a nonlinear multimode–multiproduct assignment formulation that minimizes the total generalized system cost while satisfying the usual flow conservation constraints. The modeling framework provides both an adequate representa- tion of large multimodal–multiproduct transportation sys- tems for strategic planning purposes and a mathematical structure well suited for efficient solution methods. The article also provides a framework for dealing with empty flow estimation in the rail mode. The author uses a procedure based on a gravity (or entropy maximizing) model for forecasting empty rail car traffic similar to that used for Canadian National Railways. The model uses an approach where empty cars are defined as a separate product and an OD empty movement demand matrix is estimated by the gravity model. OD demand of empty movements could be estimated by a variety of heuristic and exact methods, but a gravity model is used in this application. Cube Cargo Cube Cargo Software Package, Citilabs Inc., Oakland, Calif., 2003 [Online]. Available: http://www.citilabs.com/cargo/ index.html. Citilabs’ Cube system is a local transportation planning package. Three interrelated products comprise the system: Voyager for passenger transport, Cargo for freight transport, and ME for freight and passenger matrix estimation. Cube Cargo uses commodity-based four-step modeling to estimate freight flows and assignments. It uses OD ma- trices and input commodity data and multivariate linear re- gression models to determine tons of each commodity group produced and consumed in each zone. Each produc- tion and consumption commodity is further divided into in- ternal (local) and external (import and export) segments. Productions and consumptions are distributed by a combi- nation of short-haul and long-haul movements, each using a gravity model with different cost functions. Short trips travel by road; long-haul trips can be split between differ- ent modes or mode combinations based on time, distance, and cost in a multinomial logit equation. Trip assignments are calculated from mode and commodity group trip tables. In addition, two submodels cover the effects of major trans- portation and intermodal facilities and the traffic assign- ment effects of local delivery and non-goods-related truck traffic. Product documentation claims uses for local and re- gional but not statewide planning. The software was not available for evaluation. Cargo appears to be compatible with the procedures of the Guidebook on Statewide Travel Forecasting. Cube ME finds optimal trips matrices from a variety of data sources, including ground counts, using maximum likelihood estimation. Product literature suggests that the methodology is appropriate for estimating truck trip matrices. GENERAL TOOLS RELATED TO FREIGHT FORECASTING Trip Table Estimation from Ground Counts Value of Data Sources Rios, A., L.K. Nozik, and M.A. Turnquist, “The Value of Different Categories of Information in Estimating Freight Origin-Destination Tables” (CD-ROM), Presented at the 81st Annual Meeting of the Transportation Research Board, Jan. 13–17, 2002. Trip table estimation from ground counts or other par- tial data sources has application to freight forecasting be- cause the entire OD table for a given mode or a given com- modity is rarely known. The authors report on their ability to estimate multimodal freight OD tables from a variety of information sources, including ground counts, origin to- tals, destination totals, and flows between specific OD pairs. Tests were performed on two supernetworks con- taining rail, truck, and rail–truck intermodal. The authors varied the quantity and quality of inputs to the OD estima- tion process and compared the results with a known trip table. They found that link count data were most useful for correctly ascertaining OD tables, followed by origin and destination totals. Of all links counts, the ones on higher volume facilities had the greatest positive impact on the quality of the OD matrix. The formulation used to estimate OD tables involved minimizing an entropy term while simultaneously minimizing the squares of errors to input data. Weights were applied to each item of data in the min- imization based on its quality.

101 State of the Art in OD Matrix Estimation Abrahamsson, T., “Estimation of Origin-Destination Matri- ces Using Traffic Counts—A Literature Survey,” IR-98- 0212, International Institute for Applied Systems Analysis, Laxenburg, Austria, May 1998. OD trip table estimation from ground counts has been an ac- tive field of research for almost three decades; however, most applications of the concept are still considered experimental. A recent comprehensive literature review is provided in this doc- ument, which looks at 24 different research efforts on static trip table estimation. A number of mathematical approaches to the problem have been proposed (e.g., entropy maximization, gen- eralized least squares, linear programming, Bayesian inference, and heuristic algorithms), but there does not yet appear to be a consensus as to the best method for any given problem. All of the most researched methods require that a solution to an opti- mization problem be found, and all methods are highly com- putational. Particular attention in this review article is given to generalized least squares formulations of the optimization problem and gradient search methods for solving the required optimization problem. The formulations often require a con- siderable amount of information from a traffic assignment, typ- ically the probability that a trip measured on a given link had its origin and destination in a particular OD pair. When there is congestion on the network, these probabilities are dependent on the OD table used in the assignment and, by logical extension, the estimated OD table. Thus, many of the reviewed articles deal with the problem of obtaining stable OD tables in con- junction with equilibrium traffic assignment. Because the solu- tion of the OD estimation problem is not unique (there are usu- ally many OD tables that can be equally good at fitting a set of ground counts), a persistent line of research has incorporated a “target” OD table into the formulation, with the idea of pre- serving as much information as possible from the target. Genetic Algorithm Kim, H., S. Baek, and Y. Lim, “Origin-Destination Matrices Estimated with a Genetic Algorithm from Link Traffic Counts,” Transportation Research Record 1771, Trans- portation Research Board, National Research Council, Washington, D.C., 2001, pp. 156–163. Genetic algorithms (GAs) use many iterations and random- ized factors to “evolve” an optimized solution. Unlike gravity model or Fratar model iterative adjustments, GAs do not nec- essarily converge toward the optimal solution with each itera- tion and require much more computing power and many more iterations. GAs for developing OD tables may have the poten- tial to be robust, requiring less input data than currently needed. The paper explains GAs thoroughly, including a step-by- step method, and offers an example of building an OD table from link flow volumes. The goal is a tool to make data collection cheaper. The sample 8-node network required approximately 500 iterations to construct a trip table from link volumes and had an average error of 2.5% for each esti- mated OD pair volume. The sensitivity of the algorithm to erroneous data was measured by running the sample network with various (5%, 10%, 15%, 20%) error rates in link volumes. In each case the algorithm, once converged to a solution, had individual OD pairs of up to twice the error rate (for example, 5% underre- porting of link volume led to maximum 10% underreporting of the same link in the finished table). However, the errors averaged out throughout the network to below the original link volume error rate (e.g., 15% error on each link led to an average of 8.6% error for the entire network). This paper is a very good introduction to the concept of genetic algorithms and their applications. The methods and theory are described clearly, and the sample network is ex- plained well. A Wyoming Application Wang, J. and E.M. Wilson, “Interactive Statewide Trans- portation Planning Modeling Process,” Transportation Re- search Record 1499, Transportation Research Board, National Research Council, Washington, D.C., 1995, pp. 1–6. In Wyoming the primary concerns for transportation plan- ning are goods movements and tourism travel, especially during summer months. Consequently, the goal of the travel model is to obtain weekend truck and passenger vehicle traf- fic flows. Automatic traffic record reports, the port of entry truck counts, and the vehicle miles book data are used to build an OD trip table. An entropy maximization scheme is used to estimate a vector of Xa’s, which are used to find the number of trips between zones i and j: (D1) where Tij = number of trips from i to j, paij = proportion of trips from i to j that use link a, tij = trips from i to j from a preliminary OD table, and Xa = trip estimation factor for link a with a traffic count. A program was written in Visual Basic, with the paij’s com- ing from a Quick Response System II output file. In Wyoming, 50% of goods movements are external–external, so having a well-estimated truck trip table from ground counts is important to the planning process. Practical Issues Van Aerde, M., H. Rakha, and H. Paramahamsan, “Estima- tion of O-D Matrices: The Relationship Between Practical and Theoretical Considerations” (CD-ROM), Presented at T t Xij ij a Pija a = ∏

102 the 82nd Annual Meeting of the Transportation Research Board, Jan. 12–16, 2003. The paper is a good overview of many different OD table estimation techniques. It includes a brief discussion of syn- thetic OD estimation based on entropy, minimum informa- tion, and link counts. The authors discuss two approximation techniques. No specific model or application is presented. Surveys State-of-the-Art Review Lau, S., “Truck Travel Surveys: A Review of the Literature and State-of-the-Art,” NCHRP Web Doc 3, MTC, Oakland, Calif., Jan. 1995, In Multimodal Transportation Plan- ning Data: Final Report, 1997 [Online]. Available: http:// books.nap.edu/books/nch003/html/166.html. The paper describes different survey methods used by various metropolitan planning organizations (including Chicago; Ontario; Vancouver; Phoenix; Alameda County, California; New York–New Jersey; El Paso; and Houston– Galveston) to estimate truck trips. The paper also gives in- formation about types of data, use of truck survey data, and method used. Most of the surveys divide trucks by weight, number of axles, and truck type. The major survey method is a telephone–mailout–mailback, and the main uses of the truck data are for regional truck travel model development and corridor and route analysis. The paper discusses the cost of surveys, gives recommendations for new truck data and analysis tools, and lists numerous truck facts. Survey forms are not provided. Radwan, A.E., M. Rahman, and S.A. Kalevela, “Freight Flow and Attitudinal Survey for Arizona,” Transportation Research Record 1179, Transportation Research Board, Na- tional Research Council, Washington, D.C., 1988, pp. 16–22. A mail survey about highway freight movements and car- rier attitudes was designed by the Arizona Freight Network Analysis to evaluate the performance of Arizona’s highway system for freight movement. The survey had two parts, a freight movement survey and a carrier attitudinal survey. Af- ter analysis the existing data (including Reebie data and Associates of Greenwich, Connecticut), the Arizona Freight Network Analysis researchers found that existing data could be used, so the group decided to use a random mail survey to gather data. The survey was only sent to the freight carriers. Carriers were selected according to their rank by total annual miles: 100% of the top 1,200 carriers, 50% of carriers from 1,201 to 1,900, and 5% of carriers from 1,901 to 12,900. The survey mailing package asked carriers to provide informa- tion for a “recent representative week” of 7 days. The freight movement survey obtained the carrier code, contact person, date, carrier type, shipping date, commodity shipped, gross weight, shipment’s destination city and state, Arizona routes taken in travel, and comments. Conclusions about Arizona’s freight traffic were made. Methodology Casavant, K.L., W.R. Gillis, D. Blankenship, and C. Howard, Jr., “Survey Methodology for Collecting Freight Truck and Destination Data,” Transportation Research Record 1477, Transportation Research Board, National Research Council, Washington, D.C., 1995, pp. 7–14. The paper describes the methodology and procedures used to collect statewide freight truck data in Washington State, and the authors say that it was the first study of col- lecting statewide freight truck OD data directed through per- sonal interviews with the truck drivers. Roadside interviews of truck drivers were judged by the authors to be the most ef- ficient technique. The survey was designed to be statistically reliable for all major Washington State highways and was implemented over a 24-h period in each of the four seasons. Permanent weigh stations and ports of entry were used as the primary data collection sites. The questionnaire included in- formation on time-of-day movements, truck and trailer con- figuration, cargo type, payload, use of intermodal facilities, routes used between major origins and destinations and the specific route. Approximately 30,000 drivers were inter- viewed. Because the survey was designed to obtain statewide information for each season, it needed to be conducted for 5 weeks at 25 sites at each season, and the survey needed to maintain consistency as to day-of-week at each site. Inter- view team recruitment and training were important compo- nents of the survey method. The importance of the officer, equipment needs, and the quality control procedures are also described in the paper. There are three particular lessons from the Washington State freight truck survey. First, community service clubs can be a viable labor force for conducting personal interviews. Second, the officer is a critical factor in obtaining informa- tion from truck drivers. Last, site set up and the use of sys- tematic sampling techniques are the important factors to get traffic flows and promote cooperation at the interview sites. MODAL STUDIES Rail Demand Study Nazem, S.M., “Forecasting Rail Freight Transportation Demand,” Business Economics, Vol. 11, No. 4, 1976, pp. 65–69. The article looks at forecasting rail freight in two ways, by aggregate derived demand and by commodity derived

103 demand, and then compares the two. The aggregate derived demand approach uses a simple econometric model based on GNP. The commodity derived demand approach forecast is based on certain major commodities handled by railroads. Both models showed the same degree of performance. From the users’ point of view the commodity model is easy to vi- sualize; however, the aggregate model can provide a good monitoring system under normal economic conditions. Analysis of Carload Waybill Data Lee, H. and K. Viele, “Loglinear Models and Goodness-of- Fit Statistics for Train Waybill Data,” Journal of Transporta- tion and Statistics, Vol. 4, No. 1, 2001 [Online]. Available: http://www.bts.gov/publications/jts/v4n1/paper5/lee.html. The paper is concerned with estimating counts of carloads by commodity type and by origin and destination as found from rail waybills. Log-linear models can be used to compare flows of freight between different areas, search through data for unusual flows, and make predictions of future flows. The authors used waybills from states that have more than 4,500 carloads per year of traffic or 5% or more of a state’s traffic from 1988 to 1992. Analysis focuses on the origins of ship- ments, the destinations, and the types of commodity. STCC codes are used to divide commodities into groups. The full log-linear model in this context is log mijk = logai + logbj + logck + logdij + logeik + logfjk + loggijk (D2) or mijk = aibjckdijeikfjk gijk (D3) where ai is a main effect for origin i (and bj and ck are analo- gous); dij is an interaction effect for when origin i and desti- nation j have cargo flows not proportional to the product of the main effects ai and bj (e and f are analogous); and gijk is a three-way interaction between origin i, destination j, and commodity k. The actual counts nijk of commodity k from i to j follow a Poisson distribution with mean mijk: (D4) It was found that train cargo flow is not related to distance; therefore, the paper focuses on the complex interaction ef- fects of the variables. Rail Costs Morlok, E.K. and J.A. Warner, “Approximation Equations for Costs of Rail, Trailer-on-Flatcar, and Truck Intercity Freight Systems,” Transportation Research Record 637, Transportation Research Board, National Research Council, Washington, D.C., 1997, pp. 71–77. P n m m e n ijk nijk mijk ijkkji ( ) ( ) ! = − === ∏∏ 1 37 0 181 0 181 ∏ This paper describes and compares equations for estimat- ing the costs of rail, trailer-on-flatcar, and truck intercity freight systems. Data sources are old, but the general proce- dures are still applicable. Rail carload. For analysis purposes a typical car was assumed to have a 59-Mg (65-ton) weight capacity and a vol- ume capacity of approximately 139 m3 (4,900 ft3). The pro- cedure determines shipment characteristics (shipment weight, shipment distance, commodity density, type of rail car used, and highway-access coefficient), computes the number of rail cars that are needed for shipment, selects an applicable cost formula, and calculates cost. Other consider- ations are the costs of interchange movement between two railroads, intertrain switching within the same company, and intratrain switching of cars of the same train. The study re- gion was east of Wisconsin and Illinois and north of Ken- tucky and North Carolina. Trailer-on-flatcar. Trailer-on-flatcar costs are calculated from an assumed cost per ton-mile carried. Operation costs are based on a regional average, which is taken from infor- mation in the Rail Carload Cost Scales 1973. The procedure for estimating costs is the same as for the rail carload, using dimensions for trailer-on-flatcar trailers of V = 2,550 ft3 and W = 490 cwt. Highway common carrier. The ICC Statement and the Cost of Transporting Freight by Class 1 and 2 Motor Com- mon Carriers of General Commodities 1973 were used for estimating the costs on the highway intercity freight system. The main difference between the highway system cost- estimating procedure and the rail-carload procedure was explicitly taking into account the density of the shipment. Air Freight Carey, E., H.S. Mahmassani, and G.S. Toft, “Air Freight Usage Patterns of Technology-Based Industries,” Trans- portation Research Record 1179, Transportation Research Board, National Research Council, Washington, D.C., 1998, pp. 33–39. This is a survey of technology-based firms in the Austin, Texas, area. The survey was conducted concerning actual use of air transportation and other modes by these firms. Survey- ors wanted to determine for both inbound and outbound freight such information as origin and destination, description of items shipped, weight, size, other characteristics, approximate value, frequency of shipments, and mode of shipment. Firms were identified using a directory of manufacturers. Eight-six firms were contacted in person, and then by mail. Only 13 firms responded to the original detailed form. A shorter less detailed form was then developed. Only 20 more firms responded to the shortened form. In total, only

104 33 of 86 (38.4%) were returned and only 18 of that 33 gave information about freight. Owing to the small number of firms surveyed, only trends can be shown in various aspects of freight. This includes re- gions of shipments and mode (parcel, air, truck), as well as regions where air freight is sent to and from and where air passengers travel. Results show a high dependency of tech- nology-based firms on the parcel and air services. Truck Truck Routing Wang, X. and A. Regan, “Assignment Models for Local Truckload Trucking Problems with Stochastic Service Times and Time Window Constraints,” Transportation Research Record 1771, Transportation Research Board, National Re- search Council, Washington, D.C., 2001, pp. 61–68. The paper describes a framework for a local truck routing algorithm. Routing is by load (load A, load B, load C, etc.), not geographically. Focus is on the probabilities involved with sending a truck to pick up and deliver loads within its prescribed time windows and the incurred costs. The model is logistical, and it lacks a geographic component beyond time and cost. It may be useful for determining the effect of new technologies on truck operating costs, and may also have use in modeling industrial interactions within a metro- politan area beyond I-O analysis. Truck Weight Hewitt, J., J. Stephens, K. Smith, and N. Menuez, “Infra- structure and Economic Impacts of Changes in Truck Weight Regulations in Montana,” Transportation Research Record 1653, Transportation Research Board, National Research Council, Washington, D.C., 1999, pp. 42–51. This study, although not concerned with statewide freight forecasting directly, uses similar tools to explore two important directions. First, from commodity flows, the authors use a Regional Economic Modeling, Inc. (REMI) model to determine the public infrastructure costs of pro- jected freight flows at several potential future weight lim- its. Second, the authors use the REMI model to develop comparative economic feedback forecasts (productivity, employment, personal income, and Gross State Product) based on projected freight flows at several potential future weight limits. The model and methods are not discussed in detail. Data required are I-O tables from the U.S. Department of Commerce, Bureau of Economic Analysis and commodity- flow surveys from the U.S. Department of Commerce, Bureau of Census. The data are free. The methodology is proprietary and is not discussed in the paper. The study discussed trucks only, and did not explore the effect of mode split owing to different truck weights. It also did not explore changes in industrial efficiency (pro- ductivity, consumption, substitution, etc.) beyond changing truck weight; nor did it consider economic growth unrelated to transportation. Therefore, the results are predictable. Lower axle weights result in lower public infrastructure costs, higher private infrastructure costs, higher cost of trans- portation per unit of commodity, lower productivity, lower employment, lower personal income, and lower Gross State Product. Higher axle weights show the reverse effects. The main point of this paper is to show the additional uses of a statewide freight model to predict infrastructure costs and to feedback economic input data. Trucks Between the United States and Mexico Mendoza, A., C. Gil, and J. Trejo, “Multiproduct Network Analysis of Freight Land Transport Between Mexico and the United States,” Transportation Research Record 1653, Transportation Research Board, National Research Council, Washington, D.C., 1999, pp. 69–78. The authors use border-crossing data and all-or-nothing traffic assignment to identify rail and truck differences in cross-border trade, looking for inequalities, issues, or under- served markets. This is a large-scale model analogous to the FAF or the later Latin American Trade and Transportation Study (LATTS). The data are from the Mexican Secretariat of Commerce and Industrial Development. OD tables comprise 104 zones, including Mexican and U.S. states and zones near the border crossings. Mode split is already known, deter- mined by the data source. Traffic assignment uses the STAN program. The output indicates that all-or-nothing assignment was used. There was no attempt to forecast future growth or sys- tematic evaluation of mode split. Like FAF or LATTS, this analysis uses data and traffic assignment to provide a snap- shot of the existing (1996) freight network. Trucks and NAFTA McCrary, J. and R. Harrison, “North American Free Trade Agreement Trucks on U.S. Highway Corridors,” Trans- portation Research Record 1653, Transportation Research Board, National Research Council, Washington, D.C., 1999, pp. 79–85. The authors use border-crossing data and all-or-nothing traffic assignment to identify truck cross-border truck corri- dors within the United States. The study looks at truck traffic

105 crossing the Canadian and Mexican borders. This is a large- scale model analogous to the FAF or the later LATTS. The data are from the U.S. Department of Commerce and the U.S. Department of Transportation, Bureau of Trans- portation Statistics. Traffic assignment and network model are done on TransCAD. The output indicates that all-or-noth- ing assignment was used. There was no attempt to forecast future growth. As with FAF or LATTS, this analysis uses data and traffic assign- ment to provide a snapshot of the existing (1996) freight network. Truck Forecasting Russel, E., R. Miller, and E. Landman, “Monitoring Travel Patterns of Heavy Trucks—Summary Report,” unpublished, prepared for the Kansas Department of Transportation, K-Trans Study No. 92–93, 1997. The Kansas DOT, surprised by truck traffic growth in western Kansas, collected survey data and modeled the traf- fic assignment on their in-house network. Unlike many other modeling efforts that are driven by congestion concerns, K-Trans’ primary interest was truck axle weight (ESALs) for better pavement design. They experienced data processing problems owing to per- sonnel issues, and their results were limited to modeling the resulting OD tables. There was no attempt to forecast future truck growth. The vehicle-based survey data told planners where external trips terminated, but the state’s network lacked external zones and links to distribute the trips. After several years’ delay, the project was terminated. Special Uses Middleton, D.R., J.M. Mason, Jr., and T. Chira-Chavala, “Trip Generation for Special-Use Truck Traffic,” Trans- portation Research Record 1090, Transportation Research Board, National Research Council, Washington, D.C., 1986, pp. 8–13. Special-use truck traffic is usually very short (fewer than 100 mi), tends to be cyclic in nature (trips made several times in a typical day), and both the origin and destination are the same from month to month, but will change eventually. The Texas State Department of Highways and Public Trans- portation supported a study to obtain trip generation charac- teristics for this kind of truck traffic. This study was divided into two parts, with each looking at different kinds of com- modities. One group was called “agriculture” and included timber, grain, beef cattle, cotton, and produce; the other group was called “surface mining” and included sand, gravel, and crushed stone. The first step in the study process was se- lecting special-use industries, which have specific com- modities unique to the Texas highway system. The second step was determining industry characteristics. On-site, tele- phone, and field interviews were used at sites that represent primary operations, have a significant number of trips, or ex- hibit a widespread problem in Texas. The next step was to determine vehicle characteristics associated with selected in- dustries. This task was accomplished by telephone requests to related state departments. The last step was determining trip-making characteristics. Data for both the radius of influ- ence, representing maximum distance from a center, and trip generation rates were collected from on-site interviews. The center was selected randomly and a total of 83 were chosen. The vehicle classification count was made using 15-min in- tervals for all traffic entering and leaving the facility during a total time period of one day. Marine Inland Navigation Systems Analysis Veith, M.T. and M.S. Bronzini, “Commodity Flow and Mul- timodal Transportation Analysis for Inland Waterway Plan- ning,” Transportation Research Record 636, Transportation Research Board, National Research Council, Washington, D.C., 1997, pp. 8–14. The purpose of the paper is to describe a model and to ex- plain how it is used to estimate demand for inland waterway transportation. Forecasts of demand for commodity trans- portation are provided by commodity flow analysis and allo- cated to modes by means of multimodal analysis based on cost and performance criteria. The INSA model allows cost and level of service to in- fluence the spatial patterns, mixes, and quantities of com- modity flows. Dependent variables were developed for quan- tities shipped from region i to region j, by mode; by mode and route; and by mode, route, and network element (node or link). Simultaneous equation models and direct-demand models can be used; however, a chain of sequential models is used for the INSA. The INSA commodity-flow model uses economic activity by region and analyzes flow patterns by use of multiregional general equilibrium logic, which is similar to an I-O model. The commodity flow iterates through a series of calculations to arrive at predicted annual commodity flows for the current year and demand estimates for the following year. The model uses successive approximations to forecast commodity flows and tests for convergence between the last two iterations. The main features of the model are calculating minimum cost and location, allocating demand, estimating transportation cost, forecasting economic activity, allocating commodity flow, and computing consumption.

106 The principle output of the model is a set of region-to- region commodity flows that can be used in the planning process, and additional outputs include regional economic activity, national income, and value-added. The INSA model does not use a separate mode split model, but combines a modal share and network routing analysis. It also uses a circuity constraint of an ellipse of given eccentricity being constructed about the origin and destination regions for a particular commodity movement. In addition, an optional inertia effect may be used to constrain a specified portion of any commodity shipment to observe modal-share percentages input by the user for that shipment. Hawnn, A.F. and F.M. Sharp, “Inland Navigation Sys- tems Analysis,” Transportation Research Record 636, Transportation Research Board, National Research Council, Washington, D.C., 1977, pp. 14–22. The purpose of the INSA commodity-flow model is to forecast the demand for interregional bulk commodity trans- portation. The model is an I-O model in which market dy- namics determine the location, composition, and pricing of output, and the behavior of economic aggregates determines the level of output. The model uses economic inputs such as economic activ- ities, regional attributes for 173 BEA areas, demand, and transportation costs. Operations in the model are determina- tion of minimum cost and location, computation of consumption, determination of demand, organization of transportation costs, forecast of economic activity, and allocation of commodity flow. Outputs include a commodity- flow report, a domestic-demand report, and an origin-flow report. The operations of the path-selection algorithm yield iden- tification, number of tons assigned, shipping costs for each commodity shipment, and shipping costs and transit time of assigned traffic. An ellipse about the origin and destination is used to reduce the number of paths, and commodities may also be restricted as to which modes of transportation they may use. INSA also includes an optional inertia effect, and an iterative procedure is used to assign shipments to the network. Time–Series Analysis Branyan, C.O. and G.D. Mickle, “Projecting Commodity Movements for Inland Waterways Port Development,” Transportation Research Record 669, Transportation Re- search Board, National Research Council, Washington, D.C., 1978, pp. 5–7. This study contains a preliminary analysis including gathering information from the BEA and local agencies to identify various commodity groups and historical data. The study also contains a detailed analysis to compute a series of time–series equations for straight line, second- degree curve, exponential curve, and second-degree expo- nential curve. Owing to multicollinearity problems with the original 15 variables, 5 runs were made with different combinations of variables that limited the problems. After running mathematical methods to obtain forecasts, non- mathematical procedures based on judgment and the knowledge of analysts was used. The study was very basic in nature. Intermodal Intermodal Demand in Arkansas Ozment, J., “Demand for Intermodal Transportation in Arkansas,” Walton College of Business, University of Arkansas, Fayetteville, unpublished paper (undated, around 2001). In this white paper, the author asserts that the demand for truck and rail intermodal services in Arkansas is the result of ineffective public policy relating to intermodal and misper- ceptions of traffic managers as to the cost advantages of in- termodal. The author states that application of conventional logistics theory would suggest many additional opportunities for intermodal shipping, especially in commodities of low value per weight. The analysis applied a series of cost as- sumptions to a variety of specific commodities (not com- modity categories). The computed total logistics cost was composed of Total Cost = OC + CC + Tr + PC + It + SS + Other (D5) where OC = order placement cost, CC = inventory carrying cost, Tr = transportation cost, PC = product cost, It = inventory in transit cost, and SS = safety stock cost. and where OC = A(R/Q), CC = 1/2(QVW), Tr = rRwt/100, PC = VR, It = iVRt/365, and SS = BVW. and where Q = optimal order quantity (EOQ), Q  (2AR/VW)1/2, A = cost of placing an order,

107 R = annual rate of use, V = value per unit, W = carrying cost as a percentage of average value of inventory, r = transportation rate per 100 pounds (CWT), wt = weight per unit, i = interest rate or cost of capital, t = lead time in days, and B = buffer of inventory to prevent stockouts. Thus, TC = A(R/Q) + 1/2(QVW) + rRwt/100 + VR + iVRt/365 + BVW (D6) Original Source: Coyle, J.J., E.J. Bardi, and C.J. Langley, Jr., The Management of Business Logistics, 6th ed., West Pub- lishing, St. Paul, Minn., 1996 and Ballou, R.H., Business Logistics Management, 4th ed., Prentice Hall, Upper Saddle River, N.J., 2004. COMMODITY STUDIES Agricultural Effect of Unit Trains Linsenmeyer, D., “Effect of Unit-Train Grain Shipments on Rural Nebraska Roads,” Transportation Research Record 875, Transportation Research Board, National Research Council, Washington, D.C., 1982, pp. 60–64. This paper explores the effect of a change in market area and effect on truck ton-miles by a change in rail operating practice. Truck ton-miles increased 71%, profitable length- of-haul increased by 8–15 mi, and heavier trucks became more profitable as a result of switching from single-rail car- loads to unit-trains, with an accompanying concentration of grain elevators. The paper provides a clear methodology for relationships between agricultural production and different modes, but the relationships are not applicable outside of this case. Not di- rectly useful for modelers, the paper does provide a good ex- ample of unintended effects. Shipper Expectations of Rail Vachel, K. and J. Bitzen, “Long-Term Availability of Rail- road Services for U.S. Agriculture,” Transportation Re- search Record 1790, Transportation Research Board, National Research Council, Washington, D.C., 2002, pp. 66–72. This study is a survey of the expectations of Midwest grain shippers and railways on the effect of technology and market changes. It has limited use for models in agricultural areas, because it reflects opinion only and no specific data or analysis are presented. Consensus is that light-duty track mileage and car fleet size will decrease owing to low growth and increasing efficiency. Mileage losses will occur as heavier-duty mainline track, which can carry heavier cars, will cause elevator expansion along main routes and elimi- nate market areas of smaller elevators on other routes. Empties Comparison of Methods Holguin-Veras, J. and E. Thorson, “Practical Implications of Modeling Commercial Vehicle Empty Trips” (CD-ROM), Presented at the 82nd Annual Meeting of the Transportation Research Board, Jan. 12–16, 2003. The paper compares four methods of modeling empty truck trips for feedback into OD tables. Empty trips represent 30% to 50% of freight trips. Each method is used in a two- zone simulation and in a 26-zone simulation based on data from New York City. All methods produced undercounts of the total number of truck trips, but the Holguin–Veras and Thorson (HVT5) and Noortman and van Es (NVE) methods produced the smallest errors, approximately 5% to 6% of to- tal observed trips. Hazardous Materials Erkut, E. and T. Glickman, “Minimax Population Exposure in Routing Highway Shipments of Hazardous Materials,” Transportation Research Record 1602, Transportation Re- search Board, National Research Council, Washington, D.C., 1997, pp. 93–100. This paper uses a two-step routing method for hazardous truck shipments. First it sets a constraint criterion, such as population along a network link. Any links exceeding the cri- terion are excluded from the second step. The second step is a typical shortest path or minimum impedance routing algo- rithm. The larger implications or applications to modeling or forecasting are not explored. Applications to oversize, over- weight, or other constrained trucks are not explored. Coutinho-Rodrigues, J., J. Current, J. Climaco, and S. Ratick, “Interactive Spacial Decision-Support Systems for Multiobjective Hazardous Materials Location-Routing Prob- lems,” Transportation Research Record 1602, Transporta- tion Research Board, National Research Council, Washing- ton, D.C., 1997, pp. 101–109. The authors created a computer application, ISDSS, to model hazardous material flows, including production or generation, transport, use, and disposal or processing. The model is oriented toward risk management, not transport.

108 The application optimizes flows and locations based on user- defined networks and weighted criteria. The model algo- rithms are not discussed. This is not routing or forecasting software, but solution software for hazardous material risk management. Chang, T., L. Nozick, and M. Turnquist, “Routing Haz- ardous Materials with Stochastic, Dynamic Link Attributes: A Case Study” (CD-ROM), Transportation Research Board, National Research Council, Washington, D.C., 2002. The authors describe a multi-objective routing algorithm with case study. The label-correcting network algorithm uses a convolution-propagation approach, based on the algo- rithm’s ability to “test” different paths and rule out some early. The multiple objectives are incorporated using proba- bilities, including hazardous material release probabilities and driver and population exposure probabilities. The algo- rithm includes variables for congestion and time of day to de- termine different routes. The algorithm is clearly presented. The example network is summarized, and only a single route is determined. Patel, M.H. and A.J. Horowitz, “Optimal Routing of Hazardous Materials Considering Risk of Spill,” Trans- portation Research A, Vol. 28A, No. 2, Mar. 1994, pp. 119–132. The authors propose an algorithm for routing hazmat that minimizes the risk of population exposure to airborne toxic substances that might be released in a crash. ADVANCED METHODS STUDIES Mode Split Log-Linear and Logit Models Murthy, A.S.N. and B. Ashtakala, “Modal Split Analysis Using Logit Models,” Journal of Transportation Engineer- ing, Vol. 113, No. 5, 1987, pp. 502–519. The study includes the analysis of survey and question- naire data collected in Alberta, Canada, from shippers and consignees. Log-linear and logit models were then used to create a more statistically credible and comprehen- sive method to identify the dominant modes of commodity movement. Communities were classified as shippers (sources), con- signees (sinks), or both. Major commodity-flow data such as type, mode, loads (full or less than full), control, hire (private or for-hire), and market share were gathered, as well as de- mographic data such as population, retail sales volumes, etc., and other data from transportation and government agencies. Of the surveys gathered, 1,318 of the responses were ship- pers and 6,175 were consignees. The data from both shippers and consignees were com- bined, and the five explanatory variables for modal choice considered in the study were average shipment size, loads (full or less than full), hire (private or for hire), and control. The study only uses the truck and rail modes, because all other modes carry very little freight. For the log-linear model different combinations of load (L, i = 1,2), hire (H, j = 1,2), and mode (M, k = 1,2) were con- sidered. For example, a log-linear model [LH][MH] is a model that includes the association of “loads” and “hire” in- dividually with “modes.” Two cases yielded a likelihood ratio less than one. After testing the two for statistical signif- icance and finding the chi-squared value, the model with fewest variables was chosen because there was no substan- tial difference. The saturated model chosen was [LM][HM] and is expressed as ln mijk = μ + μL(i) + μH(j) + μM(k) + μLM(ik) + μHM(jk) (D7) The second part of the study was to construct a logit model to develop a table of log odds to understand how changes in the combined levels of explanatory variables affect the response variable. The logit equation that was used is defined as logitij = 2[μM(1) + μLM(i1) + μHM(j1)] (D8) Odds ratios then were calculated and proportions were found using the transformation proposed by Berkson in 1944 by mode, load, and hire for different commodities. Mode Choice Factors Wilson, F.R., B.G. Bisson, and K.B. Kobia, “Factors That Determine Mode Choice in the Transportation of General Freight,” Transportation Research Record 1061, Trans- portation Research Board, National Research Council, Washington, D.C., 1986, pp. 26–31. This study examines the factors that shippers in eastern Canada use to determine modes of freight shipments by hired truck, private truck, and rail. A survey was used instead of gathering waybill data, because waybill data do not include level of service attributes and differences in record keeping, and many shippers consider waybill data to be proprietary. This study classifies four factors for mode choice, includ- ing characteristics of the transportation system, characteris- tics of the shipment, characteristics of the local carriers, and characteristics of the shipper. Analysis was performed using three linear logit models. The difference between the first two is that one considers in- transit damage and one considers commodity value because

109 both cannot be used in the same model owing to multi- collinearity problems. The third model uses derived variables instead of specified variables. The data showed increasing use of rail as length of transit increases. Shipping cost, in-transit damage, and commodity value were not significant in influencing mode choice. Sig- nificant influences for both rail and private truck modes were not covered in this survey. Data suggest that model data should be gathered using personal interview data, which pro- vide a higher level of accuracy and could be used to explore other factors not covered in a survey. Flows Accuracy Study Metaxatos, P., “Accuracy of Origin-Destination Highway Freight Weight and Value Flows” (CD-ROM), Presented at the 82nd Annual Meeting of the Transportation Research Board, Jan. 12–16, 2003. This paper presents a method of estimating interstate or international (external–internal) freight flows using matrices and commodity data similar to common internal–internal OD tables. The external side of the commodity flow is repre- sented by a single data source, a seaport or border crossing. The internal side of the commodity flow is represented at the county level using existing disaggregation. A set of OD ma- trices uses a gravity model to simulate long-distance freight flows and determine value or weight. Results are determined by the number of iterations of the gravity model, which is governed by the desired confidence level. The paper does not explore possible expansion of the technique to external– external or internal–internal freight flows or to distributed external sources. No example problem or comparison to an existing gravity model is provided. Disaggregation Sivakumar, A. and C. Bhat, “Fractional Split-Distribution Model for Statewide Commodity-Flow Analysis,” Trans- portation Research Record 1790, Transportation Research Board, National Research Council, Washington, D.C., 2002, pp. 80–88. A variation of the four-step modeling process, the authors create a Texas model using fractional–split distribution. This distribution uses fractions to determine origins and destina- tions. For example, zone A produces 1 good, and zone B con- sumes 1/10th of the good. Zone B also consumes 1/20 of the same good produced at Zone C. The data were from the Reebie TRANSEARCH database. Only three commodity groups were used. Beyond the struc- ture of fractional–split methodology, the model is not shown. The model results are compared to a normal four-step grav- ity model process. The fractional–split and gravity models were in close agreement. The advantages and disadvantages of each method are not clearly discussed. Gravity Model Black, W.R., “The Utility of the Gravity Model and Estimates of Its Parameters in Commodity Flow Studies,” Proceedings of the Association of American Geographers, Vol. 3, 1971, pp. 28–32. The paper reports and evaluates the results obtained from applying the gravity model to 24 sets of interregional com- modity flows for the United States in 1967. The study uses a variation of the gravity model, substituting shipments and demands for productions and attractions. The total shipments and demands are assumed known for each region and represent the row and column sums for a commodity-flow matrix. The only unknown term used in the study was the friction factor coefficient and it was increased by 0.025 in a stepwise procedure until the correlation be- tween the actual and estimated flows failed to increase. The study used high regional generalization in the flows reported between the nine census regions and 81 possible inter- and intraregional flows. The interregional distance was defined as half the square root of the region’s area. The estimates obtained from the gravity-type trade model for the 24 shipper groups were quite accurate and the model accounted for 93% of the variance in the flows examined. Overall, the study suggests that it is clearly possible to esti- mate reliable friction factors. Ashtakala, B. and A.S.N. Murthy, “Optimized Gravity Models for Commodity Transportation,” Journal of Trans- portation Engineering, Vol. 114, No. 4, 1988, pp. 393–409. The objective of the study was to reexamine survey data (Murthy and Ashtakala 1987) and develop models for com- modity transportation. A gravity model with a new technique for calibration is proposed. The commodity data were classified and survey data about shippers and consignees were gathered. The data col- lected were origin and destination of commodity movement, type of commodity, type of firm, annual tonnage, average shipment size, type of load (full or less than full), type of hire (private or for-hire), control (yes or no), and market share. Demographic data were also gathered. OD tables were developed showing origins (sources) and destinations (sinks). A series of production-constrained grav- ity models were then applied to the data from source to sink

110 and compared differences in interchanges using regression analysis. The gravity model with the highest R2 value was used as the best representation of real data. The spatial sepa- ration factor is specific to each commodity category, so there is one gravity model for each category. The models are shown by statistical measures and commodity haul fre- quency diagrams to be acceptable. Trip Length Holguin-Veras, J. and E. Thorson, “Trip Length Distributions in Commodity-Based and Trip-Based Freight Demand Model- ing Investigation of Relationships,” Transportation Research Record 1707, Transportation Research Board, National Re- search Council, Washington, D.C., 2002, pp. 37–48. The trip length distribution (TLD) in freight-demand mod- eling can be defined as either a tonnage TLD or vehicle TLD for different models. The main aim of this paper is to exam the characteristics of the tonnage TLDs and vehicle TLDs to find the relationship between the two and to identify problems when using TLDs. The shape of a TLD will be different within different environments in which freight movements take place. Major generators have a significant impact on the shape of a TLD. If a mathematical relationship between the two types of TLDs can be found, it will help exploit the best features of commodity-based and truck-based models. Input–Output Application of I-O for Commodity Flows Sorratini, J.A., “Estimating Statewide Truck Trips Using Commodity Flows and Input-Output Coefficients,” Journal of Transportation and Statistics, Vol. 3, No. 1, Apr. 2000, pp. 53–67. Sorratini, J.A. and R.L. Smith, Jr., “Development of a Statewide Truck Trip Forecasting Model Based on Com- modity Flows and Input-Output Coefficients,” Transporta- tion Research Record 1707, Transportation Research Board, National Research Council, Washington, D.C., 2002, pp. 37–48. This study used inexpensive data, the 1993 Commodity Flow Survey and I-O coefficients to create freight trip gen- eration tables. The resulting tables were used in a standard four-step modeling process. Results were generally within 25% of traffic counts. These papers deal only with freight highway flows. The network was made of 72 internal zones within Wis- consin, plus 70 external zones. Zones matched the Reebie Associates TRANSEARCH data used for freight produc- tions. Network characteristics were not discussed. Freight internal productions consisted of the 1993 Com- modity Flow Survey, employment for each zone and economic sector (U.S. Census County Business Patterns), population for each zone (U.S. Census), and tons of com- modity per truck for each STCC (Reebie Associates). Each commodity is disaggregated into STCC, zone, and internal- to-internal or internal-to-external trip type. Truckloads were assumed to be uniform across 6 days each week (312 days per year). The final freight productions were a series of 624 tables showing tons produced by each of 28 STCC sectors at each of the 624 network TAZs in Wisconsin. Productions for external zones were not considered. Freight internal attractions were determined using I-O co- efficients. From the IMPLAN software package and 1994 Wisconsin data (source not cited) and IO coefficients (source not cited), the monetary amount of one product needed by each industry to produce its output were summed for each of the 28 sectors used, resulting in a statewide estimate of total internal freight attraction volume. The total was disaggre- gated by employment (U.S. Census County Business Pat- terns) to the TAZ level. If no reliable employment numbers were available, population was used for disaggregation. Freight external attractions (imports) were based on the IMPLAN final-demand report. Demand was disaggregated to the TAZ level using employment and/or population. IMPLAN regional purchase coefficients (source not cited) determined the amount of final demand allocated to internal and external supply. Trip distribution was done by the gravity model function of TRANPLAN. Traffic assignment was mentioned, but not discussed. The model was calibrated to 40 selected links. Root mean square error (RMSE) of predicted flows against actual counts (Wisconsin DOT) ranged from 32% to 61% under different conditions of complexity. Several iterations of the gravity model changed the RMSE range from 27% to 57%. The highest errors were in the lower volumes. Volumes of more than 2,000 vehicles and three or more gravity model itera- tions had an RMSE of 27%. Vilain, P., L. Liu, and D. Aimen, “Estimation of Commod- ity Inflows to a Substate Region,” Transportation Research Record 1653, Transportation Research Board, National Research Council, Washington, D.C., 1999, pp. 17–26. This is a method of developing external-internal trip ta- bles using existing I-O data and commodity-flow data instead of cordon counts and surveys. The article includes a calibrated example. The sum of estimated commodity flows calculated from I-O tables in the example was within 6.6% of the observed value shown in the 1993 Commodity Flow Survey. Weighted average error for individual commodities was up to 28%.

111 Data required are I-O tables from the U.S. Department of Commerce, Bureau of Economic Analysis, and commodity- flow surveys from the U.S. Department of Commerce, Bureau of Census. The advantages of this method are that the data collection is easy, including data for calibration, and that the method can be set up on existing spreadsheet applica- tions. Disadvantages include methodological assumptions and limitations that can lead to significant errors in some commodity groups. The method, using matrix algebra, is to convert the I-O table into a supply-side commodity-flow matrix. Then the method determines the location quotient (percent of regional employment divided by percent of total U.S. employment) for each industry. Commodities and industries are 38 stan- dard types used by the U.S. Department of Commerce for I-O tables. The final result is external-to-regional commodity flows for each industry and commodity type. The possibility of using areas smaller than regional (such as county or TAZ level) was not explored. Two important assumptions are that commodities used by each industry are identical nationally, and cannot vary from region to region, and that the fraction of each commodity purchased locally is identical across all industries. For ex- ample, industry A uses four tons of commodity N input for each ton of output, regardless of the local price or availabil- ity or possibility of substitution. Furthermore, if 10% of all of the state’s use of commodity N is produced internally, then industry A will purchase 10% of its N locally, regardless of production, transportation, regulatory, or other impacts. In the example problem, the sum of commodity flows was measured against the 1993 Commodity Flow Survey bench- marks. The sum of all commodities was overreported by 6.6% of actual flow, with a weighted mean average of 28% error for individual commodities. Mineral products and petroleum and coal products predictions were particularly poor, and removing them changed the sum of commodities underreported by 9% with a weighted mean average of 17% for individual commodities. Final matrix and table results from this method are yearly flows. Breakdowns by day or time-of-day were outside the scope of study. Networks and Traffic Assignment Sequential Shipper-Carrier Networks Friesz, T.L., J.A. Gottfried, and E.K. Morlok, “A Sequential Shipper-Carrier Network Model for Predicting Freight Flows,” Transportation Science, Vol. 20, No. 2, 1986, pp. 80–91. The authors report on the development of three similar in- tercity freight models that have their greatest emphasis on ob- taining accurate traffic assignments. Each of the models con- tains these major elements: trip distribution with a doubly con- strained gravity model; a shipper mode choice and route choice process; a way of allocating shipments to carrier net- works; and a carrier route choice process. Separate, but com- patible, networks are used to model the shipper and carrier routing decisions. The shipper routing decision process is based on an elastic-demand user-optimal equilibrium assign- ment, where as the carrier routing decision is a fixed-demand system optimal over the carrier’s subnetwork. These shipper and carrier decision steps are sequential (without feedback). The carrier choice networks provide for movements across carriers, backhauling, and delays along mainlines and in yards. Both link cost and link traversal time are used in the route choice process. Tests were conducted on networks with up to 15 commodity groups and up to 15,000 single-direction links (arcs). The unit of spatial aggregations was a BEA region. Hypernetworks Friesz, T.L. and E.K. Morlok, “Recent Advances in Network Modeling and Their Implications for Freight Systems Plan- ning,” Transportation Research Forum Proceedings, 1980, pp. 513–520. This paper reports on initial efforts at building freight forecasting models that are superceded by later work by the same authors. The intent of the paper is to draw a distinction between passenger models and freight models. The authors concentrate on traffic assignment and show how transship- ment can be accommodated with a hypernetwork. The paper makes two contributions: (1) it shows how user-optimal equilibrium assignments may be accomplished with multiple classes; and (2) it demonstrates that carriers can choose their own criteria for optimizing their paths. State of the Art, Early 1980s Friesz, T.L., R.L. Tobin, and P.T. Harker, “Predictive Inter- city Freight Network Models: The State of the Art,” Trans- portation Research A, Vol. 17A, No. 6, 1983, pp. 409–417. The authors review several similar approaches to intercity freight forecasting, with an emphasis on their own work. They concentrate on routing decisions within macroscopic equilibrium network frameworks. Although the authors men- tion heuristic attempts to solve complex problems of shipper and carrier behavior, they are much more interested in algo- rithms based on optimization theory or variational inequality theory and how these models have developed incrementally as additional theory is added to what has already been done. They critique six full-scale models in terms of 16 attributes: • Treatment of multiple modes, • Treatment of multiple commodities, • Sequential loading of commodities, • Simultaneous loading of commodities,

112 • Treatment of congestion phenomenon via nonlinear cost and delay functions, • Inclusion of elastic transportation demand, • Explicit treatment of shippers, • Explicit treatment of carriers, • Sequential solution of shipper and carrier submodels, • Simultaneous solution of shipper and carrier submodels, • Sequential solution of macroeconomic model and trans- portation network model, • Simultaneous solution of macroeconomic model and transportation network model, • Solution employing nonmonotonic functions, • Explicit treatment of backhauling, • Explicit treatment of blocking strategies, and • Inclusion of fleet constraints. “Blocking strategies” refers to means of collecting carloads into trains for more efficient shipping by rail. The authors report at length on recent (at that time) attempts to impart further realism to the models in the areas of simultaneous shipper-carrier decision making, including competitive and cooperative behaviors, simultaneous macroeconomic and network models, and fleet constraints. Issues that have not been handled well according to the authors are backhauling and blocking. Whole Models National Energy Model U.S. Department of Energy, Energy Information Administra- tion, Office of Integrated Analysis and Forecasting, The Na- tional Energy Modeling System: An Overview 2000, Mar. 2000 [Online]. Available: http://www.eia.doe.gov/oiaf/archive/ aeo00/overview/index.html. The National Energy Modeling System is used to forecast U.S. energy production, demand, and prices over 20 years. The model is composed of a series of modules. Each module includes a single type of supply, conversion, demand, or other input or output of the system. Modules include macroeco- nomic activity, carbon emissions, transportation demand for energy, electricity markets, oil and gas supplies, coal markets, and more. The results of the model are one input to the annual U.S. Department of Energy Annual Energy Outlook report. Each annual iteration of the forecasting model includes multiple baseline cases and changes in assumptions. For ex- ample, the model for 2000 included 5 baseline cases plus 32 nonbaseline cases to explore the impacts of varying key as- sumptions. This method of varying key assumptions is not significantly used in freight transportation forecasting. The “integrating” module expedites changing assumptions by ensuring data uniformity among modules and by testing module output for iterative convergence. The integrating module automatically relaxes some impedance parameters (prices) to encourage convergence, if needed. Each module of the model can also be executed independently. The macroeconomic module includes four submodules to predict economic activity. Rather than build a model of the economy, the Department of Energy rents four models from Standard and Poor’s/DRI: U.S. Quarterly Model of the Econ- omy, Personal Computer Model of Industrial Output, Employment Model by Industry, and Regional Model. These four models each provide some of the variables for input to the National Energy Modeling System, and provide valuable cross check on results and assumptions. The macroeconomic module integrates the results of the four models and provides limited ability to question assumptions within them to pro- vide baseline cases. State of the Art, Early 1970s Smith, P.L., “Forecasting Freight Transport Demand—The State of the Art,” The Logistics and Transportation Review, Vol. 10, No. 4, 1974. This is an excellent review essay about the evolution of six major approaches to freight-demand forecasting through the early 1970s. The author reviews 44 articles relating to: (1) market share models, (2) I-O models, (3) inventory theo- retic models, (4) gravity models, (5) abstract mode models, and (6) linear programming models. The paper focuses on the assumptions behind and limitations of each approach. The author emphasizes the tradeoff between analytical or theoretical sophistication and the amount of data necessary for calibration. The simplest and most prevalent technique is mode-share models, but gravity models and linear- programming models offer better policy sensitivity and should extrapolate better to future situations. The author cau- tioned against I-O models with advice that is still relevant: “The fundamental problem of using input-output models in multiregional or multi-country analysis is the massive data requirements. This problem would be even more severe for a modally disaggregated, transport oriented inter-regional input-output model.” State of the Art, Late 1970s and Early 1980s Winston, C., “The Demand for Freight Transportation: Model and Applications,” Transportation Research, Vol. 17A, No. 6, 1983. In what could be described as a follow-up review essay to the one by Smith (1974), Winston reviews several demand for- mulations developed by researchers through about 1981. This article was written during a period of deregulation in the U.S. freight industry; therefore, the review was slanted toward those models that would help readers understand deregulation issues. The article offers opinions as to the most worthwhile directions in freight-demand models, promoting disaggregate

113 models, particularly those with behavioral underpinnings, over aggregate models. The paper emphasizes the need to look at a full range of options that relate to the mode selection deci- sion by shippers, including shipment size, service quality, and location. The author briefly discusses “inventory” models that reflect decisions made by the receiving firm. Combined Model Chang, E., A. Ziliaskopoulos, D. Boyce, and S. Waller, “So- lution Algorithm for Combined Interregional Commodity Flow and Transportation Network Model with Link Capac- ity Constraints,” Transportation Research Record 1771, Transportation Research Board, National Research Council, Washington, D.C., 2001, pp. 114–121. This paper describes a classic regional, statewide, or in- terstate model framework. The initial data are I-O tables and transportation network, and the final output is OD demand for each node in the network, link volumes, and system cost. The model has an entropy coefficient to find cross-hauling and dispersion effects. I-O flows are converted to commodity flows and to truck- loads using different conversion factors for each commodity. The model goal is a combination of OD demands and link volumes that result in optimum system cost. Once the OD demands are set, the algorithm uses Danzig–Wolfe decom- position to distribute the flows to the network. The test network included 36 zones and 13 commodities. The algorithm functioned as expected, but was not validated or calibrated. The model recognizes congestion and capacity constraints, but not truck weight constraints or time-of-day issues. Use of Secondary Data Sources Chin, S., J. Hopson, and H. Hwang, “Estimating State-Level Truck Activities in America,” Journal of Transportation and Statistics, Jan. 1998, pp. 63–74. This study estimates the amount of freight shipped by truck within, to, from, and through each state. The data come from the 1993 Commodity Flow Survey, the 1992 Census of Agriculture, the 1992 Truck Inventory and Use Survey, the 1993 and 1994 Transborder Surface Freight data, the 1993 U.S. Waterway Data, and the 1993 county business patterns. Truck flows were assigned to the Oak Ridge National Highway Network. Assignment was based on shortest path, with a travel time impedance factor. Agricultural trips were estimated and added. Import and export freight was estimated and added. Assignment of orig- inating, destination, and through flow was adjusted for inter- national imports and exports, with the port shown as “through” rather than origin or destination. The ton-mile estimate error may be up to 7% off, owing to the disparate data sources. This study used a simple model and simplifying assump- tions for the network. It did not go through the four-step model, instead loading tons, origins, and destinations into ton-miles on the network. The goal was a set of ton-mile es- timates, not assigned trips. Feedback to Generation Park, M. and R. Smith, “Development of a Statewide Truck- Travel Demand Model with Limited Origin-Destination Sur- vey Data,” Transportation Research Record 1602, Trans- portation Research Board, National Research Council, Washington, D.C., 1997, pp. 14–21. The authors explore a method of creating statewide OD tables using very limited initial data and a selected-link- based (SELINK) analysis. This method, applied to a statewide model using OD data from only 14 of 624 zones, underreported trips by only 18%. The goal is to provide a tool to lower the cost of data. SELINK analysis is a feedback process from traffic as- signment back to trip generation. The entire trip generation- to-gravity model-to-traffic assignment and then feedback to trip generation process requires three iterations to provide best results. Each selected link is compared with known vol- umes after traffic assignment, and an adjustment is com- puted. For the statewide model example, there are 32 selected links. Details of the methods, algorithms, and data requirements are clearly shown in the paper. The study covered internal– internal trips only. Error is measured by RMSE. In the statewide model example, RMSE for Interstate highways is 24%, for U.S. highways 46%, and for state highways 104%. State-of-the-Art Review Pendyala, R., V. Shankar, and R. McCullough, “Freight Travel Demand Modeling: Synthesis of Approaches and De- velopment of a Framework,” Transportation Research Record 1725, Transportation Research Board, National Re- search Council, Washington, D.C., 2000, pp. 9–16. The article offers a very good review of recent and his- torical trends up to 1999, and then develops a conceptual framework for freight transportation planning. The authors briefly review freight forecasting and data requirements, adding nothing new.

114 This is a good introduction to the field or a primer on the subject. It is a good brief summary of work in the field. Sequential Models Ashtakala, B. and A.S.N. Murthy, “Sequential Models to De- termine Intercity Commodity Transportation Demand,” Trans- portation Research A, Vol. 27A, No. 5, 1993, pp. 373–382. The objective of the study is to determine the demand for commodity transportation using the conventional sequential modeling approach. The first three stages are commodity pro- duction and consumption, distribution, and modal split. The route assignment stage is not included because the conventional all-or-nothing assignment is not found to be adequate for pre- dicting commodity transport volumes on the highway network. Survey data were gathered and log-linear and logit models were developed for modal split and an optimized gravity model was developed for distribution. Commodity demands were rep- resented graphically in the form of commodity-flow diagrams between origins and destinations. The diagrams are similar to desire line diagrams and show demands for rail and truck. From the study it is evident from commodity-flow dia- grams that the nearest source supplies the necessary com- modities to the communities around it and as the distance increases, the amount of interchange between the sources and sinks diminishes. The study shows that sequential modeling can be applied effectively for estimating commodity flows. The gravity model is also found to be applicable for various commodity categories. The modal split model used in the study is useful and innovative. Lastly, the source nearest a sink supplies the necessary commodities to it. State-of-the-Practice Review, Early 1980s Bronzini, M.S., “Evolution of a Multimodal Freight Trans- portation Network Model,” Proceedings Transportation Research Forum, Vol. 21, 1980, pp. 475–485. The paper describes the development of different national multimodal freight transportation network models that oc- curred at different times and with different visions. All are associated with the national network project, which encom- passed rail, highway, waterways, and pipeline networks developed in late 1960s. The INSA project developed a model that could determine the lowest cost path by considering both the shipment cost and the cost of delay as perceived by shippers. Node and link char- acteristics, which are related to the time and cost in the net- work, are described. The model was applied to a system that contains waterways and railroads. The model used the com- modity flows between BEA regions, consisting of both local and interregional traffic. The model’s estimates of major trends and patterns in transportation cost and traffic levels are reasonable, although local traffic estimates were not accurate. The next vision was the Transportation Systems Center Freight Energy Model, which allowed modal choice and routing decisions to be based on the energy consumption. The Transportation Systems Center model extensively re- vised the network and operation database. The links and nodes in the network were modified, and the transit time, energy use, and cost data were reestimated. The National Energy Transportation Study transportation network model expanded the study area and modified the network database. Equilibrium-seeking traffic assignment routines were developed for the study and were used to pre- dict flows on the model network. The effect of the equilib- rium assignment is described in the paper. The Electric Power Research Institute (EPRI) model fo- cused on the network effects of energy supply. The railroad routing algorithm developed by EPRI was much more detailed than before. Results from the EPRI model are not reported in the paper. Across the different visions, the greatest need for a trans- portation network model is a comprehensive interregional commodity-flow database. Cost is the most important poten- tial source of error in the modal choice and routing algorithms. Underdeveloped Regions Jones, P.S. and G.P. Sharp, “Multi-Mode Intercity Freight Transportation Planning for Underdeveloped Regions,” TTR, P523 (incomplete reference). This paper describes a freight model for parts of eight states between Brunswick on the Georgia coast to Kansas City—a corridor that is approximately 1,200 mi long and 100 mi wide. The transportation system there includes several Interstate, secondary rail lines, and waterways. The Standard Industrial Classification codes are used to describe com- modity groups and there are separated arcs in the network for highway, rail, and water modes. This is a conventional model consisting of 111 zones and 53 commodity and industry groups. For mode split, transport time and cost are indepen- dently derived from the network. Modifying a Four-Step Model Kim, T.J. and J.J. Hinkle, “Model for Statewide Freight Transportation Planning,” Transportation Research Record 889, Transportation Research Board, National Research Council, Washington, D.C., 1982, pp. 15–19.

115 The authors developed a multicommodity, multimodal statewide freight transportation planning model by modify- ing the existing Urban Transportation Planning System (UTPS) package developed by FHWA and the Urban Mass Transportation Administration. There are five classes of submodels: network analysis, freight transport demand analysis, vehicle requirements, assignment, and evaluation. Freight transport demand analysis was done in four steps: freight volume generation, interzonal commodity distri- bution, modal split, and freight volume assignment. UTPS.ULOGIT and UTPS.AGM were used in the calibra- tion of modal split and commodity distribution from freight volume OD data. Truck backhaul was estimated from the volume to be carried, the distance, truck size, cost, and OD table. A separate program dealt with empty rail car move- ments. UTPS.UROAD was used to assign trucks and cars to different networks. An Early Application in Florida Middendorf, D.P., M. Jelavich, and R.H. Ellis, “Develop- ment and Application of Statewide, Multimodal Freight Forecasting Procedures for Florida,” Transportation Research Record 889, Transportation Research Board, National Research Council, Washington, D.C., 1982, pp. 7–14. This paper documents an early effort to create a statewide freight forecasting model for Florida. The general method was OD table factoring and assignment. Belgium Van Herbruggen, B., In-Depth Description of the Tremove Model, Transport & Mobility Leuven; Leuven, Belgium, Mar. 2002 [Online]. Available: http://www.tmleuven.be/ Expertise/Download/Tremove_Description.pdf. The TREMOVE model is a Belgian model to forecast emissions. It is used to model changes in policy and tech- nology on air pollution, and is not suited for forecasting freight. Freight demand is based on mode, price, and time of day. Freight supply is based on price of vehicle and price of fuel. There is no network, no distinction between freight types, and no infrastructure. Interestingly, there is time-of-day sen- sitivity and multiple modes. Sweden Swedish Institute for Transport and Communication Analysis (SIKA), “A Conceptual Framework for Analysis and Model Support for Swedish Studies of Freight Transport and Trans- port Policy—An Idea Study,” Nov. 2001 [Online]. Available: http://www.sika-institute.se/utgivning/sam01_1.pdf. SIKA is the Swedish transportation statistics bureau. This study for a model framework draws mostly from the QRFM. The paper is an exploration of how to make the QRFM framework work with existing Swedish statistical reports and software. The model begins with economic assumptions from the Ministry of Finance, analogous to the U.S. Department of Commerce, data on employment, and manufacturing value. Matrix estimation uses employment disaggregated to zonal level. Similarly, through the four-step process, the authors explore local parallel data sources and software to stay close to the QRFM method. Other Morlok, E. and S. Riddle, “Estimating the Capacity of Freight Transportation Systems,” Transportation Research Record 1653, Transportation Research Board, National Research Council, Washington, D.C., 1999, pp. 1–8. The authors present a method of measuring the capacity of an entire system, rather than individual links or compo- nents. Given a network with known capacities of individual components, plus known traffic patterns (OD pattern), plus fleet size, the 13 equations of the algorithm will estimate the system capacity. The system capacity can be compared with the existing flows. Additionally, a modified method can be used to estimate capacity change resulting from change in the network or fleet size. The authors used a very small rail network for their ex- ample. Applications or potential to forecasting or modeling are not discussed. The 13 equations are shown. Don Breazeale and Associates, Inc., “Task II—Data Col- lection Strategic Analysis Report for Strategic Planning Advice for Freight/Truck Model Development Project, Pre- pared for Los Angeles County Metropolitan Transportation Authority, Oct. 2002. At 234 pages plus Executive Summary, the report covers only Task II (data collection strategies) of Los Angeles County MTA’s regional Freight Forecasting project. It does not include the model. The report has a useful summary of data sources, methods, and technologies, some of which are useful for statewide forecasting. No new methods are devel- oped. The consultant recommends long-term relationships with major shippers as a source of reliable OD data. Also in- cludes an annotated bibliography of data sources for regional and statewide modeling. Very complete and usable as a ref- erence guide.

116 A Typology Souleyrette, R., T.H. Maze, T. Strauss, D. Preissig, and A.G. Smadi, “Freight Planning Typology,” Transportation Re- search Record 1613, Transportation Research Board, National Research Council, Washington, D.C., 1998, pp. 12–19. Most models built for freight transportation are based on two concepts: spatial price equilibrium and network equilib- rium. Most of these models have had implementation diffi- culties that have limited their use. The authors contend that it is more important to focus on the economic sectors for the freight traffic demand because most state and regional eco- nomics are dominated by a few sectors. The freight planning “typology” focuses on addressing the needs of state and re- gional transportation planning. The first step is to identify key issues. Freight is divided into groups with the same trans- portation requirements. Each commodity or sector becomes a layer. Sectors are overlayed to form an aggregate forecast of all freight traffic volumes. The paper used a case study of “meat product and farm machinery industries in Iowa” to demonstrate the method.

Next: Abbreviations used without definitions in TRB publications »
Statewide Travel Forecasting Models Get This Book
×
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

TRB’s National Cooperative Highway Research Program (NCHRP) Synthesis 358: Statewide Travel Forecasting Models examines statewide travel forecasting models designed to address planning needs and provide forecasts for statewide transportation, including passenger vehicle and freight movements. The report explores the types and purposes of models being used, integration of state and urban models, data requirements, computer needs, resources (including time, funding, training, and staff), limitations, and overall benefits. The report includes five case studies, two that focus on passenger components, two on freight components, and one on both passenger and freight.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!