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.
103 This chapter describes the current practices, both domestic and international, of FG and FTG modeling. The different modeling applications of FG and FTG modeling are classified (Fischer and Han 2001) as follows: ⢠Planning applications: The main goal here is to produce estimates of FG/FTG for conglomerations of usersâ typically defined by a zoning systemâfor transportation planning purposes at the state, regional, corridor, or urban level. Typically, these are medium- and long-term studies aimed at answering questions about medium/long-term capacity needs and economic development. ⢠Engineering applications: These analyses are intended to provide key input to a variety of engineering design ques- tions concerning facility design issues, traffic operation studies, site impact analyses, provision of on/off-street parking for trucks, etc. In some cases, the analysis could focus on a single establishment, a single location with mul- tiple establishments, or an entire area, such as a downtown area. These studies emphasize short-term analyses and improvements. State-of-Practice of Transportation Planning Applications Describing the current practices of FG/FTG modeling requires contending with the multitude of modeling possi- bilities, and the lack of an accepted modeling standard. In order to simplify the problem, the team decided to focus on the specific functions that each model component is expected to performâas opposed to focusing on the modeling tech- niques themselves. Doing this provides a clear and succinct way to discuss the role of FG/FTG modeling in the context of planning applications. Figure 17 shows the main outputs of the different components of freight demand models used in transportation planning. As shown in the figure, there are multiple pathsârepresented by the arrowsâthat could be taken that reflect the options available to the analyst. It is important to highlight that each of these decisions have implications in subsequent steps. For example, deciding to model vehicle trips will lead to a situation in which freight mode/vehicle choice cannot be considered for the simple rea- son that the vehicle trips are themselves already the output of a freight mode/vehicle choice that already took place. Simi- larly, while producing an economic forecast of employment could be mapped into the corresponding land uses, typically the reverse cannot be done because land use is a compos- ite of disparate industry sectors. Data collection efforts also constrain the kind of models that could be developed. For example, if commodity flow data are not collected or avail- able, then the only alternatives are the vehicle-trip formula- tions shown in the right side of the figure. It should be said that, although Figure 17 represents the entire freight demand modeling process, the main empha- sis here is on those aspects that concern FG/FTG analyses, which are represented by the dotted box in the figure. In this context, the FG/FTG literature is discussed and classified with respect to the: ⢠Dependent variable: This is the output of the modeling effort, which could be commodity tonnage (C); vehicle trips (V); or a mix of commodity tonnage and vehicle trips (C&V). The latter represents cases where an internal truck- trip origin-destination matrix is estimated for internal- internal trips, and a commodity-based model is used for external-internal and internal-external trips. ⢠Independent variables: These are the variables used to explain FG or FTG. They typically are employment, popu- lation, land use, etc. ⢠Level of aggregation: This refers to the level of detail used in the model. Aggregate (A) models quantify the FG and FTG of a conglomeration of users while Disaggregate (D) models study the FG and the FTG patterns of individual establishments. A p p e n d i x e Description of Practice, Evaluation Criteria, and Evaluation Process
104 ⢠Level of geography: Five cases are considered: Facility spe- cific (F); Corridor (C); Metropolitan (M); Regional (R); and State (S). ⢠Data sources used: The input used in the modeling effort. Review of Domestic Practice The review is based on previous publications and addi- tional modeling applications identified by the team. This review, while by no means comprehensive, provides a solid overview of the state of domestic practice on FG and FTG planning applications. A summary of the models is presented in Table 70. To facilitate interpretation and analysis, the bulk of the modeling applications are at the state level (46%), and metropolitan areas (44%). Corridor and facility specific applications represent the remaining 10%. It is important to highlight that Table 70 only contains the applications for planning purposes, which explains the low number of facility specific examples as most of them are done for engineering applications (discussed in the next chapter). In terms of the dependent variable: 46% use vehicle trips; 49% use commodity tonnage; and 5% use a combination of vehicle trips (usually for internal-internal trips), and com- modity tonnage (for the rest of the flows). About 33% of the models are aggregated, 36% are disaggregated, and others (31%) could not be identified from the review. The indepen- dent variables used include: employment by industry sector (62% of cases); population (36%); land use variables (5%); and other variables (23%). As for modeling techniques, 33% use regression analysis, 23% use IO analysis, and 13% use FTG rates. These three modeling methods constitute the majority of the FTG models used in practice (about 69%). Other methods include cross classification (5%), matrix esti- mation (8%), and time series analysis (1%). No information could be obtained for the remaining 17%. The bulk of the models reported in the literature are linear (16 out of 18). Traffic assignment Feedback/validation process to ensure equilibrium and soundness Loaded vehicle- trips Freight mode / vehicle choice Empty vehicle- trips Commodity production and consumption patterns Commodity distribution Trip interchan- ges (O-D) Delivery tours Demand data collected/available Vehicle-trip data Commodity and vehicle-trip data Economic forecasts Vehicle-trip origin and destination patterns Vehicle-trip distribution Future land use by class Future employ- ment by sector Focus of NCFRP 25 Figure 17. Schematic of modeling pathways.
105 Table 70. Summary of modeling applications. Models t ne dn ep e D se lb ai ra V Independent Variables -e rg g A no it ag -a rg oe G yh p Modeling Technique er ut cu rt S Data Sources Heavy Truck Freight Model for Florida Ports V Month index, exported and imported freight units A F Time series, regression N Gate/vessel/container data, gantry crane data, and performance reports Cross-Cascades Corridor Model V Employment by sector D C IO L State population survey, Census data, Transearch Minnesota Truck Highway 10 Truck Trip Forecasting Model V Employment by sector D C FTG rates L Truck data, Industrial employment, Quick Response Freight Manual New York Metropolitan Region Model (Freight tunnel) C Truck payload factors by commodity group D C Transearch Atlanta Commercial Vehicle and Truck Models V Industrial, retail, office employment, households A M Regression L Truck survey data and counts Baltimore Metropolitan Council Models V Industrial, retail, office employment, households A M Regression L Truck survey, borrowed FTG rates Bangor Area Model V Employment, population, industrial productivity A M L Industrial productivity gains, Quick Response Freight Manual Chicago Commercial Vehicle Travel Demand Model V Employment by sector A M FTG rates L Commercial Vehicle Survey, trip diaries Delaware Valley Regional Planning Commission Model V D M Cross- classification L Socioeconomic data, truck survey Denver Regional Model V M Trip diaries, intercept surveys, and automatic vehicle counts Greater BuffaloâNiagara Regional Model V Employment, population, land use D M Land use FTG rates L Trip diaries, land use at trip end, carriers survey Los Angeles Regional Cube Cargo Model (2004) C Employment A M IO, regression L Socioeconomic, truck/commodity flow, and port data, intermodal-warehouse survey Maricopa Association of Governments Model V Population, employment by land use category D M Land use FTG rates L Truck survey, external vehicle trip survey, traffic counts. New York Best Practice Model V A M OD estimation Employment data, highway and transit counts, regional trip generation rates Portland Commodity Flow Tactical Model System C A M Truck counts taken around truck terminals and reload facilities Puget Sound Region Freight Model C Tonnage shipment rates by commodity, employment D M County business patterns data, SAIC/ Transmode data, national IO tables Skagit Countywide Model C D M Local economic data, surveys Southern California Association of Governments Heavy Truck Model V & C* Land uses/Industry types, employment, household D M IO, Cross classification L Shipper-receiver survey, IO tables, socioeconomic data, WIM data Vancouver Truck Freight Model V Population, employment, special generator D M Expanded originâdestination surveys, cargo volumes San Francisco Bay Area Freight Model V Employment A M Regression L Business survey, external intercept surveys, trip diaries Southeast Michigan's Regional Freight System V Total/employment acres, basic, retail, wholesale employment, households A M FTG rates, regression L Travel Survey Data Connecticut Model C S Transearch Florida Intermodal Statewide Highway Freight Model C Employment by industry, population D S Regression L Census of population, Transearch Florida Model (2001) C Employment, population D S Regression Transearch, payload data, IO table Indiana Commodity Transport Model (1993) C Employment, population D S Regression L Socioeconomic data, employment and population forecasts, county business patterns data Indiana Commodity Transport Model (1997) C Employment by industry, payloads, average percent empty by truck type S IO, regression Commodity survey, commodity groups, mail data, truck payload data Kansas Model C S Local agricultural production data Kentucky Model C S OD estimation Transearch, truck traffic counts Massachusetts Statewide Model C Truck payload factor S Commodity flow survey, truck payload data Michigan Statewide Truck Travel Demand Forecasting Model C Employment S IO L Employment data by industry, national IO tables (continued on next page)
106 Table 71 summarizes the features of the various model- ing approaches by level of geography. As shown, the bulk of the metropolitan-level planning applications use vehicle trips as the dependent variable, while statewide applications favor commodity flows. This undoubtedly reflects the inherent dif- ficulties of collecting commodity flow data in urban areas, and the presence of the Transearch database at the core of the statewide models based on commodity flows. In all other respects, metropolitan and state FG/FTG applications are quite similar. (Not much should be made of the apparent dif- ference in the aggregation level used, because of underreport- ing.) The table also shows that facility specific and corridor applications are in the minority. Review of International Practice Table 72 shows a summary of international modeling applications, mostly from Europe. Table 73 presents a sum- mary of the models by level of geography. The table shows that most models are based on economic principles (e.g., IO, multi-regional IO, general equilibrium, SCGE). These tech- niques comprise 13 out of 23 applications. They are followed by regression models that capture 10 out of the 23 cases listed. It is interesting to note that none of the models listed use FG/ FTG rates. Equally significant is that a sizable portion of the application listed is based on freight data collected through surveys. This stands in contrast with the United States, where freight data collection activities have dwindled since the 1980s. State-of-Practice of Transportation Engineering Applications This chapter discusses the application of FG/FTG to support transportation engineering applications (e.g., traffic impact fee assessment, traffic operations studies, site impact analy- sis, street design, etc., as summarized in NCHRP Synthesis Note: F denotes Facility specific applications; C stands for Corridor, M for metropolitan, and S for State. pi rt el ci he V y ti do m mo C wo lf C & V no it an ib mo C tn e my ol p m E e su d na L se lb ai ra v no it al up oP sr eh t O et ag er gg A et ag er gg as i D se ta r G TF -s so r C no it ac if is sa lc no is se rg e R se ir es e mi T OI x ir ta M n oi ta mi ts e ra en i L ra en il -n o N F & C 4 3 1 2 2 1 3 1 1 1 1 2 1 M 17 12 4 1 12 2 8 3 9 7 4 2 5 2 1 11 S 18 3 14 1 10 6 4 3 4 7 6 2 3 1 Total 39 18 19 2 24 2 14 9 13 14 5 2 13 1 9 3 16 2 er ut cu rt S Independent Variables Dependent Variables n oi ta ge rg g A le ve L Modeling Technique y hp ar go e G se sa c f o re b mu N Table 71. Modeling applications by level of geography. Table 70. (Continued). Models t ne dn ep e D se lb ai ra V Independent Variables -e rg g A no it ag yh pa rg oe G Modeling Technique er ut cu rt S Data Sources Multimodal Freight Forecasts for Wisconsin (1996) C A S IO Transearch, Census of manufacturers, value of shipment data New Jersey Statewide Model Truck Trip Table V Employment, households, truck terminals, special generators A S Regression N Commodity flows and survey data, truck counts data, Census Bureau data New South Wales Australia Model C S IO Commercial vehicle and economic survey North Carolina Model V Employment S trip diary surveys Ohio Model V Employment S Regression Oregon Model C S IO Surveys Virginia Freight Model V & C* Employment by industry group, population A S OD estimation Transearch, IO tables, employment, population Wisconsin Statewide Freight Component (2006) C Employment by industry, population, trip generators D S Regression Transearch, national IO table, zonal employment and population Wisconsin Statewide Truck Trip Forecasting Model (2000) C S IO Transearch, commodity flow survey
Models t ne dn ep e D se lb ai ra v Independent variables no it ag er gg A yh pa rg oe G Modeling technique er ut cu rt S Data Sources Ãresund Traffic Forecast Model - Denmark C&V A C Regression N Trip survey, national statistics Fehmarnbelt Model - Denmark C&V A C Regression L Trip survey, national statistics TREMOVE - Europe C&V Production, consumption, import, expenditures, value density A EU General equilibrium N National statistical data on traffic, weight- to-vehicle conversion factors STREAMS, SCENES - Europe C Production, consumption, import, export, investment, public expenditures, value density A EU Multi-regional Input Output (MRIO) N IO National tables (Eurostat) ASTRA - Europe C Production, consumption, import, export, investment, public A EU IO, system dynamics N IO National tables (Eurostat) TRANS-TOOLS - Europe V GDP per sector A EU Regression L Basematrix, traffic counts, Eurostat and EC growth projections. EUFRANET (rail) - Europe C Employment, GVA, trade imports and exports A EU Regression L Shippers survey CGEUROPE - Europe C Production, consumption, import, export, investment, public material expenditures, value density A EU Spatial Computa- ble General Equili- brium (SCGE) N Harmonized National IO tables SASI - Europe C Production, consumption, import, export, investment, public material expenditures, value density A EU Regional production functions N Harmonized National IO tables CROW Trip rate Parameters - Netherlands V Land use, number of employees D F Regression L Company surveys Hamburg Freight model- Germany V Land use, number of employees D M Regression L Company surveys FRETURB - France V Land use, number of employees D M Regression, tour generation L Company surveys Tokyo urban freight model - Japan V Land use, number of employees D M Regression, tour generation N Company surveys SAMGODS - Sweden C GDP, distances, commodity price, domestic public sector demand A N MRIO L National account data (IO) and foreign trade data, regional economic aggregates SMILE - Netherlands C Production, consumption, import, export, investment, public expenditures, value density A N MRIO, added trip gen through warehouses N National account data (make/use) and foreign trade data, regional economic aggregates export, investment, public GDP, employment, population GDP, current trips Table 72. Summary of international modeling applications. (continued on next page)
Models t ne dn ep e D se lb ai ra v Independent variables no it ag er gg A yh pa rg oe G Modeling technique er ut cu rt S Data Sources EUNET2.0 - UK C Production, consumption, import, export, investment, public expenditures, value density A N MRIO, added trip gen through warehouses N National account data (IO) and foreign trade data, regional economic aggregates. WFTM - Belgium C Production, consumption, import, export, investment, public expenditures, value density A N MRIO, elastic trade coefficients L National account data (IO) and foreign trade data, regional economic aggregates SISD - Italy C Production, consumption, import, export, investment, public material expenditures, value density A N MRIO, elastic trade coefficients L National account data (IO) and foreign trade data, regional survey data, regional economic aggregates MOBILEC - Netherlands C Production, consumption, import, export, investment, public material expenditures, value density A N Regional production functions N National account data (IO) and foreign trade data, regional survey data, regional economic aggregates RAEM - Netherlands C Production, consumption, import, export, investment, public material expenditures, value density A N SCGE N National account data (IO) and foreign trade data, regional survey data, regional economic aggregates NEMO/PINGO - Norway C Production, consumption, import, export, investment, public material expenditures, value density A N SCGE N National account data (IO) and foreign trade data, regional survey data, regional economic aggregates INTERLOG - Germany C&V Regional and sectoral economic aggregates D N Regression, Monte Carlo simulation, tour structures N National and regional economic aggregates Modelo Nacional de Transporte de Carga - Colombia C Production, GDP by department, population A N Four step, regression models N Population, economic statistics, traffic counts, origin-destination surveys Table 72. (Continued).
109 The trip rates listed in the ITE Manual are for all vehicle- trips. For some of the land use types, the ITE Manual provides an estimated percentage of truck trips among all vehicle-trips. Table 74 provides a summary of these land use types, and their share of truck trips. The techniques described in the ITE Man- ual could also be used for freight traffic. How to select inde- pendent variables and trip generation methods is described in the ITE Trip Generation Handbook, 2nd Edition (2004). The ITE Trip Generation Handbook, 2nd Edition (2004) provides guidelines for selecting independent variables, time period of analysis, and estimation methods to use. Although these guidelines are developed for all vehicle types, they could be applied to FTG. However caution should be applied when using them. The ITE Handbook also recommends the use of independent variables that: (1) appear to be a âcauseâ for the variation in trip ends generated by land use; (2) could be obtained through primary measurement and not derived from secondary data; (3) produce a rate/equation that best fits the data; and (4) are related to the land use type and not solely to the individual site characteristics of the site tenants. In terms of preferred time period of analysis, the ITE Hand- book suggests â. . . the time period in which the combination of site-generated traffic and adjacent street traffic is at its maximum.â To select the most appropriate estimation meth- ods among the graphic plot (local data collection), weighted average rate, and regression equation, a detailed step-by-step guideline is provided. The document also suggests guidelines and procedures for how to: conduct a trip generation study to obtain local generation rate; consider pass-by, primary, and diverted linked trips; estimate trip generation for general land uses; and conduct trip generation for multi-use development. Report 298). A unique feature of these applications is that they require relatively accurate estimates of FTG for a wide range of land use types to determine user fees, traffic mitiga- tion measures, among other things. These analyses are usually made using the Institute of Transportation Engineers (ITE) Trip Generation Manual (2008) and Chapter 5 of the Quick Response Freight Manual. Both the ITE Manual and the Quick Response Freight Manual describe data requirements and pro- cedures for conducting these analyses. The following sections provide a summary of these important references. ITE Trip Generation Methods This section discusses two key publications produced by the ITE: the ITE Trip Generation Manual and the ITE Trip Gen- eration Handbook. Because of their importance, some level of detail is provided in this review. The ITE Trip Generation Manual, 8th Edition (2008) contains FTG data for 162 land uses under ten major land use categories (i.e., Port and Ter- minal, Industrial/Agricultural, Residential, Lodging, Recre- ational, Institutional, Medical, Office, Retail, and Services) for several time analysis periods (e.g., weekdays and weekends). For each land use, the ITE Manual provides the weighted average trip rate, a regression equation (if there are sufficient data for estimation), and the data plot showing the observa- tions. The independent variable is typically a measurable and predictable unit describing the study site that can be used to predict the number of trips or trip ends. Typical independent variables include the number of employees, gross floor area, number of vehicles, etc. Reported statistics include average trip rate, standard deviation, and the statistics for regression analysis. Note: F denotes Facility specific applications; C stands for Corridor, M for metropolitan, N for National, and EU for Europe. pi rt el ci he V wo lf yt id o m mo C no it an ib mo C C & V tn e my ol p m E se lb ai ra v es u dn a L no it al up oP sr eh t O et ag er gg A et ag er gg as i D sn oi tc nu f no it cu do rP mu ir bi li uq E no is se rg e R OI ra en i L ra en il -n o N F 1 1 1 1 1 1 1 M 5 3 2 4 3 1 1 2 3 5 3 2 N 10 9 1 1 9 9 1 1 2 2 5 3 7 EU 7 1 5 1 1 9 7 1 2 2 2 2 5 Total 23 5 14 4 6 4 2 19 18 5 2 4 10 7 9 14 Modeling Technique er ut cu rt S y hp ar go e G se sa c f o re b mu N Dependent Variables Independent Variables n oi ta ge rg g A le ve L Table 73. International modeling applications by level of geography.
110 site analysis related to freight. The Quick Response Freight Manual was developed for the Travel Model Improvement Program by the FHWA of the United States Department of Transportation (USDOT). The Quick Response Freight Man- ual summarizes the purpose of freight-related site analysis is â. . . to estimate, within an acceptable level of accuracy, the number of new commercial trips generated by a new or planned facility and determine whether or not the existing network of primary highways, local roads, municipal streets and other transportation facilities can sufficiently handle the projected traffic demands.â The report makes a distinction between site analysis for existing and planned facilities. The recommended process for site analysis is shown in Figure 18. Freight trip generation (FTG) is thus the third step in this process. In terms of data gathering, the report specifies that data can be obtained from various sources including the developer, designer, owner, or contractor, or the local/ municipal/city engineerâs office that issues construction per- mits and approves plans and specifications. In summary, the following data items may need to be collected: ⢠Company/owner name and address. ⢠Type of facility to be operated on site (e.g., retail, indus- trial, manufacturing, warehousing, etc.) and the activities involved. ⢠Size of the facility in terms of land area, floor area, number of employees, etc. ⢠Type of commodities, products or services produced and consumed. ⢠Anticipated volume of shipments and receipts expressed in either weight, volume, dollar value or other freight units. Appendix A of the ITE Handbook (2004) specifically dis- cusses FTG. The discussion, however, is only informational and â. . . provides no recommended practices, procedures, or guide- lines.â The Appendix defines three categories of uses of FTG: (1) traffic operations that are directly affected by the presence of trucks in the traffic stream; (2) design considerations that need to be addressed with the aid of truck traffic data includ- ing both pavement design and geometric design of the street or roadway; and (3) public and political concerns about the traffic impacts of developments, that are often debated in pub- lic hearings/meetings and the press. This Appendix includes 23 references on previous research on truck trip generation, some of which contains specific trip generation data. How- ever, the Appendix explicitly states that âmany of the existing data sources are quite dated . . .â and âmany of the studies on truck trip generation have used very general categories of land use.â Therefore these existing FTG rates data should be used with caution. The Appendix also identifies the incon- sistencies in existing FTG studies in terms of the definition of trucks and truck trips, and states that: âthe independent variables that provide the greatest statistical reliability for âall- vehicleâ trip generation may not be the most appropriate for estimating truck trip generation.â Therefore, the independent variables need to be enhanced or refined, so that the most appropriate variables (predictors) can be identified and used for FTG. Quick Response Freight Manual Chapter 5 in the Quick Response Freight Manual describes specific data requirements and procedures for conducting Land Use Code Land Use Catergory Truck trips 38% of weekday traffic at container terminals 60% of weekday traffic at break-bulk terminals 021 Commercial Airport Less than 1% of weekday and weekend traffic 022 General Aviation Airport 3-5% of weekday traffic 70% of site-generated driveway volume at an intermodal 34% of driveway volume at truck terminal located on 130 Industrial Park 1-22% of weekday traffic. Average was approx. 8% 150 Warehousing 20% of weekday traffic at one of the sites surveyed 151 Mini-Warehouse 2-15% of weekday traffic at sites surveyed 152 High-Cube Warehouse 9-29% of peak hour traffic. Additional data provided 254 Assisted Living % of trucks in different time periods provided in a table 731 State Motor Vehicles Department 0.44% of the weekday traffic (range of 0.12% to 0.85%) 732 United States Post Office 1.2% of the weekday traffic 760 Research and Development Center 1.84% of weekday traffic (range of 0.4% to 4.0%) 813 Free-Standing Discount Superstore % of trucks in different time periods provided in a table 815 Free-Standing Discount Store 2% of weekday traffic 816 Hardware/Paint Store 1-3% of weekday traffic. Average approximately 2%. 860 Wholesale Market 30% of total traffic at the site 890 Furniture Store 1-13% of weekday traffic. Average approximately 5%. 931 Quality Restaurant 1-4% of weekday traffic. Average approximately 1.6%. Waterport/Marine Terminal010 Truck Terminal030 Table 74. Land use classifications with truck traffic in trip generation.
111 the new facility may also be used. It is worthy of mention that, in addition to the total number of new trips, the analyst may also be interested in the distribution of these trips on a given day, week, or even month. These temporal characteristics are important in determining the impacts of the new traffic on the peaking patterns around the site. In summary, to conduct the estimation, the following factors need to be considered: ⢠Land use. ⢠The number of employees and households. ⢠The total floor/building area, or total land area of the facility. ⢠Type, weight and volume of commodities produced and consumed. ⢠Commodity classifications. ⢠The sizes and capacities of vehicles. ⢠Modes and carriers available. ⢠The frequency and scheduling of shipments. ⢠The storage and handling operations. ⢠Other factors that influence the total demand for freight transportation by the facility. The report recommends that the analyst explore these and the many other types of relationships between anticipated freight traffic and the site/facility characteristics. Appendix D of the report contains tables that summarize the detailed daily trip generation rates for each location, land use type, and truck classification. The trip generation rates are provided for the following land use types for different types of vehicles (the numbers in the parentheses are the SIC codes): ⢠Agriculture, Mining and Construction (1â19). ⢠Manufacturing, Transportation/Communications/Utili- ties, and Wholesale Trade (20â51). ⢠Type of vehicles or carriers to be used for transportation as well as the company or agency that will be responsible for shipping. ⢠Locations of markets for commodities and services pro- duced (e.g., local, intercity, out-of-state, international). ⢠Locations of markets for materials, commodities or services used (e.g., local, intercity, out-of-state, international). ⢠Locations of intermediate facilities (i.e., warehouses, con- solidation points) that will serve the new facility. ⢠Schedule of shipping operations. The report indicates that detailed dataâsuch as the type and volume of commodities used and produced, the loca- tions of origins and destinations of the shipments, and the schedulesâneed to be obtained by conducting interviews and surveys with the appropriate individuals. In identifying the network of transportation facilities, the Quick Response Freight Manual suggests that all existing phys- ical and operational characteristics of network facilities have to be described according to size, capacity, traffic volumes, geometry, speed limits and any other restrictions on use or access (e.g., truck size and weight limits). The characteristics of the traffic that the facilities serve need to be obtained as well, because they may also be relevant to the site analysis. The sources of transportation network and traffic data include the Design and Traffic Divisions of City or Local Governments, DOTs, MPOs, and other planning agencies. The Quick Response Freight Manual highlights that for site analysis, more accurate and detailed FTG estimation is required. Two FTG methods are recommended, including: (1) site-specific trip generation rates; and (2) regression equa- tions, although other methods that can significantly improve the forecasts of the demands for freight transportation due to Source: Cambridge Systematics Inc. (1996) Data gathering - Obtain land use and economic activity data related to the planned sites Identify the network that will serve the traffic generated by the site Freight trip generation - Predict the number of new freight trips Trip distribution - Determine the origins and destinations of the new freight trips Trip assignment â Assign the freight trips to the network Determine the changes of level of service due to the new trips Figure 18. Recommended process for site analysis.
112 FTG. This reflects the fact that FG and FTG are determined by fundamentally different processes. In the case of FG, the amount of cargo generated is a reflection of a production process in which a set of intermediate inputs, labor, and capi- tal interact to create a set of economic outputs. The function of land use is, in most cases, to provide the physical space and conditions to make such production process feasible. The exceptions are specific economic activities, e.g., agriculture, in which land could definitively be considered an economic input. As a result, the ability of land use variables to explain and predict FG and FTG depends on how well a land use class could act as a proxy for the kind and intensity of the eco- nomic activity being performed at the site. In contrast, FTG is determined by the minimization of the total logistic costs associated with the transportation of the FG. In essence, once the output of the production process has been determined, businesses determine the best combi- nation of shipment sizes and delivery frequency to transport the freight generated at the site. These decisions determine the FTG. As a result, FTG is a reflection of the way the indus- try arranges itself to transport the FG. The second aspect of importance is related to the interac- tions between the level of detail in land use, and the level of aggregation used in the models. It is useful to classify the dif- ferent possibilities as a function of the level of detail used to define the land use classes, and the level of detail used in the FG/FTG modeling process. In a simplified fashion, one could define the following classification: ⢠Land use level of detail: â Specific, i.e., when the land use class maps directly into a well-defined industry sector. ⢠Retail Trade (52â59). ⢠Offices and Services (60â88). ⢠Unclassified (89). For each of the land uses, FTG rates are given in terms of employees, 1,000âs of square feet of office space, and acre- age. In the case of special trip generators such as intermodal terminals, trip generation estimates can be obtained through direct contacts with a limited number of firms and with spe- cific questions. Actual trip generation data can generally be obtained through direct contacts, observation, or surveys. If not, the report suggests the use of the default values in Appen- dix D of the report. The report describes the types of data that may be sought for different modes including highway, water, rail, and air, as depicted in Table 75. The specialized database section in Appendix K-4 of the report also includes data sources for other modes including pipelines, coal move- ments, military transportation, Mexican and Canadian trade, imports and exports, and other topics. Identification of Key Variables The identification of the variables that influence FG/FTG must take into account the inherent differences between the generation of freight, and the generation of freight vehicle trips as well as the interplay between the level of detail in the defini- tion of the land use classes, and the level of aggregation of the FG/FTG modeling effort. These aspects are discussed in this sec- tion, in connection with the identification of the key variables. As discussed throughout this document, and specifically in Chapter Three, from the conceptual and practical points of view it is best to explicitly consider the process of FG, and Source: Cambridge Systematics Inc. (1996) Special Trip Generators Data Collection Requirements Data Sources Highway Average daily freight activity per site by truck classification, for both inbound and outbound Fleet manager of the planned facility Water Loadings and unloading in twenty foot equivalent units (TEUs), or forty foot equivalent units (FEUs) Port Facilities Inventory U.S. Waterborne Exports and Outbound Intransit Shipments and the converse for imports Tonnage for Selected United States Ports Rail Origins and destinations, number of cars, tons, length of haul, Participa- ting railroads and interchange locations Carload Waybill Sample Air Freight express and mail traffic carried byairport and airline; operational information on Air and Expedited Motor Carriers Conferen- ce members by airport The Airport Activity Statistics of Certificated Route Air Carriers; The Air and Expedited Motor Carriers Network Guide and the Express Carriers Association Service Directory (produced by the Film, Air and Package Carriers Conference) Table 75. Data collection for special trip generators.
113 eling at specific land use classes could perform as well as economic variables such as employment. This is certainly the case when variables such as square footage are used to explain and predict FTG (Bartlett and Newton 1982; Tadi and Balbach 1994). The reason is related to the fact thatâat the establishment level and for the same type of economic activityâvariables such as square footage may indeed be able to capture the effects of business size on FG/FTG. In essence, for the same line of business, an establishment with a larger built area may be expected to produce more FTG than a smaller one. The combined effect of a narrowly defined land use class, together with establishment level land use data, and disaggregate FG/FTG modeling, is expected to provide sufficient results. General Land Use Classes Combined with Disaggregate FG/FTG Modeling In this scenario, a broadly defined land use class is used to group a set of business activities under a heading such as âretail,â âcommercial,â and others like them. As a result, these general land use classes tend to include rather disparate mixes of economic activities, with diverging FG/FTG patterns. The net impact is that the ability of land use variables, e.g., built area, to explain FG/FTG is reduced by the internal heterogene- ity of the FG/FTG patterns inside the land use class. The chief implication is that the larger the degree of heterogeneity, the less capable land use variables will be to explain FG/FTG. Specific Land Use Classes Combined with Aggregate FG/FTG Modeling In this scenario, a narrowly defined land use class (e.g., food stores, retail) is used in combination with aggregate modeling intended to produce FG/FTG for conglomerations of business establishments, e.g., at the zonal level. The accuracy of these approaches depends on two different factors. The first relates to how well the independent variables explain FG/FTG. The â General, i.e., when the land use class includes multiple industry sectors. ⢠Level of aggregation of the FG/FTG modeling process: â Disaggregated, i.e., when the models focus on the esti- mation of FG/FTG for a specific establishment. â Aggregated, i.e., when the FG/FTG models focus on the generation of conglomerate of businesses. It should also be said that, although there are multiple gra- dations of these groups, using such a simplified classification serves the purpose of illustrating the interplay between level of detail in land use, and level of aggregation in the model- ing process. The key combinations are shown in Table 76. It is important to highlight that Table 76 does not make a dis- tinction between the estimation of a model and its applica- tion, either for engineering or planning applications. It is also understood that the data required to estimate the models has to be consistent with the level of detail in land use, and the level of aggregation of the FG/FTG modeling effort. In other words approaches based on disaggregate modeling require disaggre- gate data as an input. Similarly, once a set of land use classes has been selected and used for model estimation, any ensuing applications will necessitate a comparable set of inputs. As shown in Table 76, four possibilities are defined. As indicated in the remarks in each cell, the different combina- tions exhibit radically different levels of performance. Specific Land Use Classes Combined with Disaggregate FG/FTG Modeling This case corresponds to the situation in which a relatively narrowly defined land use class, e.g., food retail, is used as the mechanism to group a set of establishments for modeling purposes. During the estimation process, disaggregate FG/ FTG models are estimated with the land use variables, e.g., square footage, corresponding to the establishments. The limited experience with these types of approaches indicates that using land use variables for FG/FTG mod- Disaggregated Aggregated ssal C es U dna L ni liate D fo leve L larene G cificepS Level of Aggregation in FG/FTG Modeling Process The ability of these models to explain FG/FTG diminishes with the heterogeneity of the industry sectors included in the land use class, and the explanatory power of land use variables The ability of these models to provide accurate estimates of FG/FTG could be hampered by both the validity of the aggregation procedure and the broadly defined land use classes Land use variables expected to work relatively well in explaining FG/FTG Could work well as long as the explanatory variables and aggregation procedures used are appropriate Table 76. Scenarios of land use detail and level of aggregation in FG/FTG modeling.
114 ⢠Great care must be taken to ensure that the aggregation pro- cedures used to estimate FG/FTG at an aggregate level are adequate. Not doing so may lead to large estimation errors. Criteria to Determine Best Practices The analyses conducted by the team indicate that the identification of best practices has to take into account three separate aspects: ⢠Level of detail used in the definition of the land use classes used. ⢠Level of aggregation of the FG/FTG modeling effort. ⢠Modeling technique used to actually estimate the FG/FTG model. It is important to discuss these separately because the ability of a modeling technique, e.g., regression analysis, to produce good results depends on both level of detail in land use and the level of aggregation of the model. The discus- sion takes place in two parts. The first focuses on the first two aspects, and the second on the third. The expected perfor- mance of the various combinations is assessed according to the following evaluation criteria: ⢠Data requirements for both model estimation and calibration. ⢠Conceptual validity of the approach, i.e., the consistency of the approach with respect to the reality being modeled. ⢠Practicality, i.e., how easy it is to apply the model in an application context. ⢠Relevancy to the application context. ⢠Expected accuracy of the approach, i.e., how likely it is to produce accurate results. Level of Detail in Definition of Land Use Classes vs. Level of Aggregation in FG/FTG Model Table 77 shows the teamâs assessment of the performance of the various possibilities of level of detail and level of aggrega- tion, according to the various evaluation criteria. In terms of conceptual validity and accuracy, there is no doubt that disag- gregate models are better than their aggregate counterparts. This is an obvious consequence of their enhanced ability to capture the connections between FG/FTG and the establish- ment attributes. Disaggregate models also require less data than aggregate models. It suffices to say that most FG/FTG disaggregate models are estimated with 20â50 observations, while the estimation of aggregate models typically requires the kind of data collected by origin-destination surveys. However, disaggregate models require disaggregate fore- casts, which may strain the forecasting ability of most transpor- second is much less obvious, as it relates to the ability of the model to capture the aggregate patterns of FG/FTG. In essence, a good aggregate model is one that is consis- tent with the disaggregate behavior of those users the model intends to represent, and which, as a result, is able to accu- rately predict the aggregate FG/FTG. As discussed in this final report, there is only one way to aggregate results for a given disaggregate model. If the specification used in the aggregate model is consistent with the correct form of aggregation, the aggregate model would have a better chance of making accu- rate predictions. Otherwise, it will lead to erroneous results. This is a major concern because it is routinely assumed that FTG rates are constant, when in fact, the sparse data show that a significant amount of industry sectors exhibit constant FTG, which leads to FTG rates that decline with business size. As a result, estimating the aggregate FTG as the multiplication of the total employment by a constant FTG rate is extremely problematic if the underlying FTG is constant at the establish- ment level. For instance, if the business in a given transporta- tion analysis zone has a constant FTG (e.g., five truck trips per establishment per day) the correct way to estimate the total FTG is to multiply the number of businesses by the unit FTG of five truck trips/establishment. In this context, attempting to estimate the total FTG by multiplying the total employ- ment times an assumed FTG rate per employee will translate into large estimation errors. The issue here is that the constant FTG translates into FTG rates that decrease with business size, which are not captured by the constant FTG rate. General Land Use Classes Combined with Aggregate FG/FTG Modeling This scenario considers the use of a general land use class, e.g., âcommercial,â that groups various industry sectors, in combination with an aggregate FG/FTG modeling effort. As may be expected, the effectiveness of these modeling approaches is hampered by both the heterogeneity of the FG/ FTG patterns included in the land use class, and the aggrega- tion issues discussed in the previous section. As a result, these modeling applications are expected to exhibit the lowest per- formance in terms of accuracy and conceptual validity. The chief implications of this discussion are that: ⢠FG is likely to be explained with the use of economic vari- ables, e.g., employment level, as they represent the input factors of the corresponding economic processes. ⢠FTG is determined by total logistic cost considerations. Since the corresponding explanatory variables are not readily avail- able, the estimation of FTG models is likely to require the use of proxy variables such as land use type. ⢠The performance of land use variables depends on how they are defined and integrated into the FG/FTG modeling process.
115 techniques of wider applicability. More specifically, the dis- cussion considers the cases shown in Table 78. As shown, the alternative modeling possibilities have been placed in the larger context of the modeling plat- forms in use for freight demand modeling (i.e., vehicle- trip-based, commodity-based) which makes it easier to connect FG/FTG analyses to advantages and disadvantages of these platforms. As discussed elsewhere (Ogden 1992; HolguÃn-Veras and Thorson 2000), the use of vehicle-trip or commodity-based models has a number of implications, summarized in Table 79. As the reader can see, there is con- sistency between the findings discussed in the report and the pros and cons for the modeling platforms previously identified in the literature. It is important to mention that since the modeling tech- niques, as implied in Table 78, can be applied to either mod- eling platform, they could be discussed in general terms. (The exception is IO analysis that cannot be applied in vehicle-trip models.) This discussion is presented in the following sections. FG/FTG (Constant) Rates This technique is, without any doubt, the simplest and most widely used of all. The reasons are obvious, as the FG/FTG tation demand models. This may require the use of assump- tions to allow for the extrapolation based on the observed conditions for the base year. For these reasons, it is fair to con- clude that disaggregate models are slightly less practical than aggregate ones. In terms of appropriateness for engineering and planning applications, here again disaggregate models are the best alternative. As indicated in the table, while disag- gregate models could meet the needs of both types of applica- tions, aggregate models, because of their nature, could only deal with planning applications. The picture that emerges is that disaggregate models do pro- vide the best overall alternative, though there are issuesâsuch as the ones concerning the need for disaggregate forecastsâ that need to be dealt with. In spite of this limitation, the team is confident that disaggregate FG/FTG models do represent the most promising approach, particularly if the CFS micro-data are used in their estimation. Modeling Techniques The second aspect of importance is the modeling tech- nique used to estimate the FG/FTG model. Although there are a number of special use (mostly facility specific) tech- niques that could be used (e.g., time series, artificial neural networks), the main emphasis of this discussion is on the Note: Better (+++) to Worse (---) gnireenign E snoitacilppa gninnal P snoitacilppa Specific Disaggregate Economic (+++) (+++) (++) (++) (+++) (+++) Land use (+++) (+++) (++) (++) (+++) (+++) Specific Aggregate Economic (++) (++) (---) (+++) (---) (++) Land use (-) (++) (---) (+++) (---) (++) General Disaggregate Economic (-) (--) (+++) (++) (++) (--) Land use (--) (--) (+++) (++) (++) (--) General Aggregate Economic (--) (---) (--) (++) (--) (--) Land use (---) (---) (--) (++) (--) (--) Model features: Appropriateness for: L ev el o f de ta il on la nd ssalc esu noitagergga fo leve L troffe gniledo m fo tnednepedni fo epy T desu elbairav Evaluation criteria: stne meriuqer ata D ytidilav lautpecno C ytilacitcar P ycarucc A Table 77. Summary evaluation of combinations of land use detail and level of aggregation. Vehicle-trip based Commodity based Rates FTG rates FG rates Regression FTG regression models FG regression models Input-Output Not applicable IO models Modeling platform Modeling technique Table 78. Cases considered.
116 direct proportionality between FTG and the independent variable used, will underestimate the FTG for businesses smaller than average, and will overestimate the FTG for those larger than average; as shown in Figure 13. It is not clear how well a constant rate would perform in modeling FG as no such applications have been reported in the literature. However, the team would expect that since the problems that impact their application for FTG mod- eling (e.g., indivisibility of vehicle trips, the role of ship- ment size) do not impact FG, a constant FG rate could work reasonably well. FG/FTG Regression Models In these techniques, a statistical relation between a depen- dent and a set of independent variables is empirically estab- rates are computed directly from the data, typically as the summation of the vehicle trips (FTG) or tons (FG) produced and/or attracted by a sample of establishments, divided by the summation of the values of a suitable independent vari- able (e.g., employment, gross area). Although in almost all cases, vehicle trips are used, there is no reason preventing the use of rates to estimate commodity generation. Graphically, the rate is the slope of the line through the origin in Figure 19. Although pragmatic and simple to use, the use of a con- stant FTG rate has a number of problems: ⢠It forces the estimated FTG to pass through the origin, without statistically testing the validity of such assumption (as discussed, for instance, 85% of the industry sectors in NYC do not meet this assumption). ⢠As a consequence of the above, using a constant FTG rate in industry sectors that do not meet the assumption of y = 0.0233x + 3.3992 y = 0.0975x 0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 10.0 0 20 40 60 80 100 N um be r o f d el iv er ie s Number of employees Deliveries The FTG for establishment below average size will be underestimated The FTG for establishment below average size will be overestimated Figure 19. Constant FTG rate model vs. regression model with intercept. Vehicle-trip based Commodity based Independent variable (vehicle-trip) is easy to measure Does consider the economic characteristics of the cargo It includes loaded and empty trips Resembles the real life process Does not consider the economic characteristics of the cargo Requires surveys to estimate commodity flows Poor connection to the underlying economics Requires complementary models to estimate loaded and empty trips Cannot be used for mode choice Modeling platform Impacts Disadvantages Advantages Table 79. Implications of alternative modeling platforms.
117 The use of IO models is generally considered to be a solid technique that is grounded in well-established economic the- ory. However, there are a number of observations that should be made: ⢠The set of parameters that link the output to the corre- sponding inputs, i.e., the technical coefficient matrix, is usually estimated using national or regional accounting techniques. For that reason, their validity is doubtful for smaller geographic units such as transportation analysis zones (Bureau of Economic Analysis 2007). ⢠The fixed nature of the technical coefficients (variable coef- ficients have been found extremely difficult to estimate) does not allow IO models to consider structural changes in the economy that may change the proportions in which the inputs are consumed. For these, and other reasons the use of IO models has been phased out in Europe in favor of general equilibrium models, which are able to capture the connections between economic activities, regional patterns of commerce, and transportation accessibility. These models estimate FG as an endogenous variable, i.e., as an output of the model, and have been suc- cessfully applied at the regional and European levels. Taken together, the teamâs assessment of the various model- ing techniques can be summarized in Table 80. As shown in the table, constant rates, though practical and not requiring much data, are of questionable validity and likely inaccurate, because of the embedded assumption of proportionality between FTG and business size. Regression models, particularly at the disag- gregate level, do seem to represent the best technique available for both engineering and planning applications. IO modeling, though conceptually solid, cannot address the needs of engi- neering applications that require estimates of FG/FTG. lished. Of great significance is that the statistical signifi- cance of the independent variables is assessed, and that, as a result, the final models only contain the independent vari- ables that play a meaningful role in explaining the phenom- enon under study. These techniques are extremely flexible, and are able to estimate linear and nonlinear models, with or without intercept, and under several assumptions of cor- relation structures. Regression models do tend to require more data than FG/FTG rates; while most analysts would feel comfortable estimating simple regression models with about 30 observations, FG/FTG rates could be estimated with a handful of data points (and even with only one). With such a combination of flexibility, and statistical rigor in the inclusion of only the independent variables that do play a role, it is hard to argue against the validity of FG/FTG regression models. Input-Output Models These techniques are based on economic tabulations of the cost of inputs (e.g., steel, energy) that are required to produce a unit of economic output (e.g., a certain number of cars) at the national or regional level. Once the economic output, or final demand, is set, the IO matrix can be used to esti- mate the corresponding inputs, also referred to as interme- diate demands. In the case of multi-region IO applications, where the origin of the inputs must also be estimated, these are accomplished with the assistance of spatial interaction models. The commodity flows are estimated from the eco- nomic flows using suitable values of cargo, and reversing the origins and destinations (the commodity flows run counter to the economic flows). For reviews of freight demand IO modeling the reader is referred elsewhere (Kanafani 1983; de la Barra 1989). Note: Better (+++) to Worse (---) gnireenign E snoitacilppa gninnal P snoitacilppa Constant Rates FG (+) (+) (+++) (+++) (-) (++) FTG (--) (---) (+++) (+++) (--) (--) Regression FG (+++) (+++) (+) (++) (++) (+++) FTG (+++) (++) (+) (++) (++) (+++) Input-Output FG (+++) (++) (--) (--) (--) (++) FTG n.a. n.a. n.a. n.a. n.a. n.a. Model features: Appropriateness for: euqinhcet gniledo M ledo m fo epy T Evaluation criteria: stne meriuqer ata D ytidilav lautpecno C ytilacitcar P ycarucc A Table 80. Summary evaluation of modeling techniques.
