Freight-Demand Modeling to Support Public-Sector Decision Making (2010)

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.

72.1 Overview of Outreach Efforts The approach to this study was driven by a desire to under- stand the needs of decisionmakers and planners and to assess the degree to which existing technical tools meet these needs. This is a departure from the more traditional method of reviewing existing models and determining the most feasible improvements. As a result, this approach relied heavily on information collected from a series of interviews conducted concurrently with freight modelers, planners, and decision- makers from state DOTs and MPOs as shown in Figure 2.1. These in-person and phone interviews were supported with a Web survey and a comprehensive review of freight demand forecasting literature that focused on models and analysis tools that enhance the understanding of freight demand and public-sector decisionmaking. Taken together, these outreach efforts allowed the development of an under- standing of the types of tools currently used in practice, the key issues and challenges faced by practitioners, and the types of improvements to freight modeling capabilities that would be of the most interest to practitioners. 2.2 Public-Sector Freight Analysis Needs and Available Tools Public-sector agencies have a wide range of freight analy- sis needs, including the following: • Costs and benefits of freight programs and projects, • Performance measures specific to freight movements, • Mode shifts in response to program and policy changes, • Time-of-day shifts in response to program and policy changes, • Route diversion estimates in response to program and policy changes, • Freight forecasts in response to program and policy changes, • Existing routings of freight vehicles, and • Facility flow information at important freight handling facilities. As shown in Figure 2.2, public-sector freight planners, mod- elers, and decisionmakers place a premium on information describing existing freight routing; and also require informa- tion on freight costs and benefits and flows at individual freight facilities. To address this broad range of freight analysis needs, public agencies often require multiple freight analysis tools because a single model application is typically not appropriate to address all statewide or regional needs. As such, public agen- cies require a suite of models and analytical tools to apply to different problems and to address the questions asked by deci- sionmakers. Typical freight analysis tools, including the unique advantages and disadvantages of each, are described in the fol- lowing subsections. Time Series Models Based on historical or observed data over a period of time, time series models provide short-, medium-, and long-term forecasts of freight demand and freight activity. Time series– based forecasts range in sophistication from simple regression modeling established on past freight activity levels to more complex multivariate autoregressive models. Examples of time series freight model development and implementation are common among public-sector agencies. Tools commonly are available as trend analysis tools in many commercial com- puter analysis packages, including the widely used Excel. With high-frequency data, time series methods produce good short-term forecasts and require less time and fewer resources to develop than other modeling approaches. How- ever, building a proper time series model requires a long series of observed data. Also, time series models assume that the underlying economic conditions on which the forecast is based remain the same throughout the duration of the time series data and continue forward through the forecast. As C H A P T E R 2 Current Needs and Practices

8Freight Costs and Benefits Performance Measures Estimate of Mode Shifting Estimate of Time-of-Day Shifting Route Diversion Estimates Freight Forecasts Existing Routings Facility Flow Information 3.5 3.3 3.4 2.9 3.2 3.4 3.9 3.5 1 32 4 5 Figure 2.1. Interview locations. Figure 2.2. Public-sector freight analysis needs.

such, time series models are unable to account for changes to market factors, freight logistics, pricing, or policy that result in freight demand fluctuations. Behavioral Models Behavioral models, which include both choice- and survey- based demand models, capture how freight shippers perceive and select from the many available freight shipment choices. The models aim to depict the complex interactions between producers, shippers, carriers, and receivers that drive freight demand. Behavioral interactions, however, are usually com- mercially sensitive and difficult to observe. Behavioral data can be collected from shipper and carrier surveys; however, conducting behavior surveys can be prohibitively expensive. The traditional four-step modeling approach has difficulty capturing the factors that influence shipper and carrier behav- ior. Although more common for forecasting passenger travel demand, examples of freight behavioral modeling remain rel- atively limited. Given the proprietary nature of private-sector decision making, understanding and modeling the logistics decisions that affect freight demand at a regional or statewide level remains a challenge for many public-sector agencies. Commodity-Based and Input-Output (IO) Models Commodity-based and IO models estimate current and forecasted freight traffic generation and distribution by linking economic activity to associated commodity flows. These mod- els use economic data and IO tables to estimate the quantity of each commodity produced and consumed in a geographic area. The commodity flows are then converted to trucks or other freight vehicle trips using average payloads or more elab- orate empty trip models. This modeling approach is well suited to representing the economic mechanisms that drive freight movements to and from manufacturers and movements into or out of a region. Similarly, elasticity in the models provides the ability to evaluate the effects of freight policy. Commodity- based and IO models, however, are not capable of capturing empty trips that factor into freight logistics decisions without the use of complementary empty trip models. The data required to develop a commodity-based/IO model include existing and forecasted commodity flow data, traffic counts, employment data, characteristics of major freight generators, forecasts of economic activity, and techni- cal coefficients to extrapolate existing production and trade patterns into the future. Although the data needs are exten- sive, generally these multimodal freight and economic activ- ity data are readily available. Examples of commodity-based and IO model development and implementation are com- mon among public-sector agencies. However, commodity- based and IO models do not account for many local truck moves, including trips from warehouses and distribution centers, fleet repositioning, empty return trips, and truck drayage moves, as well as service, utility, and construction trucks. Many of these missed truck trips are short trips within urban areas. Therefore, truck models based exclusively on commodity flow data tend to underestimate truck trips in the urban area. In addition, the commodity flow data generally are not available at the Traffic Analysis Zone (TAZ) level, and techniques of questionable accuracy must be used to dis- aggregate county-level data. Developing and implementing techniques to support analysis of finer geographic scales, empty vehicle usage, nonfreight truck trips, labor and vehicle productivity, and chaining of freight trips as vehicles or cargo may improve the functionality of existing models. Multimodal Network Models Multimodal network models forecast and optimize mode and route choice decisions for a specific OD combination based on various transportation cost attributes. They assign commodity flows to the mode (or combination of modes) and specific route within a network that minimizes total transport costs, taking into account the location of activities within the network. The models also are capable of estimating mode and route sensitivity to various cost factors. The link-node network representations of freight supply generated by multimodal net- work models are useful for determining travel times, costs, reli- ability, and overall level of service. There are few examples of public-sector model implemen- tation of multimodal network models. Although many public- sector freight models include a truck component to truck freight, few models include fully multimodal capabilities, because in many cases these models are designed to evaluate private-sector investments and operations, rather than those for the public sector. Microsimulation and Agent-Based Models Microsimulation models depict the individual movement of large numbers of units and their attributes, while agent-based modeling defines potential actors in freight transportation and an allowable set of actions and interactions. The models allow agencies to perform “what-if” analysis and study the behavior of a system without building it. The New York Metropolitan Transportation Council (NYMTC) uses a microsimulation approach to estimate travel patterns for their regional best practice model. Although microsimulation and agent-based model development and implementation are common among public-sector agencies, they are data intensive and expensive to build. In the absence of sufficient supporting data, modelers must make many distribution assumptions to build the mod- 9

els, which may or may not reflect how freight routing decisions are made by private-sector operators, shippers, and receivers, and this contributes to challenges in interpreting results. Supply Chain and Logistics Models Supply chain and logistics models aim to capture the upstream and downstream relationships between suppliers and customers and the decisions that drive freight demand. They actually estimate the total logistics cost of shipping, including direct transportation expense and inventory cost associated with modal lot sizes and service profiles. The mod- els assume that customers (shippers) select the lowest-cost option, and they depend on information about logistical fac- tors in transportation and industry. Shipments are assigned to one mode or another, while allowing for uncertainty associ- ated with inventory risk, carrier performance, or unmeasured factors. These models can help provide information on a number of topics that would be of interest to public-sector freight plan- ners, particularly freight trip chaining and mode-choice deci- sions. However, most of these models were initially developed with the intention of helping producers (who ship goods) decide on the best choices among shipping options. Usefulness of these models for more general transportation planning is highly dependent on the actual availability of modal service options for the specific type of commodity being shipped and the shipper’s specific set of customer destinations. Without that information, such models can overstate opportunities for modal diversion due to inability to sufficiently filter out modal options that are not really available. Network Design Models Network design models include private-sector models for locating factories, distribution centers, warehouses, and other freight-generating facilities. Freight logistics companies and freight carriers must consider the frequency, mode, routing, and scheduling of freight movement within a network to pro- vide high-quality, low-cost, reliable service to their customers. Network design planning is very challenging given its scale, complexities, and decision interdependencies. Likewise, net- work design formulations are very difficult to solve, except in the simplest of scenarios. As network design models inherently relate to private- sector operations and efficiencies, examples of public-sector model implementations or applications remain scarce. Given the proprietary nature of the data required to build and opti- mize a network design model, the public sector faces obsta- cles to applying network design techniques for their decision- making purposes. Routing and Scheduling Models Typically used by the private-sector freight community, routing and scheduling models optimize the routing and fre- quency of shipments. The objective of these models is to min- imize vehicles, vehicle miles traveled, and labor; satisfy service requirements; maximize orders; and/or maximize the freight volume delivered per mile. Different types of routing and scheduling models solve problems that range in complexity as follows: • Traveling salesman problem—Determines the shortest path routing through a tour of destinations, visiting each destination exactly once and returning to the starting origin. • Vehicle routing problem—Allocates vehicles and assigns routings from a central location to serve a set of geograph- ically dispersed customers while minimizing the total dis- tance traveled. • Vehicle routing problem with time windows—Schedules and allocates vehicles and assigns routings from a central location to serve a set of geographically dispersed customers with time-window requirements. • Pickup and delivery problem with time windows— Determines vehicle assignments, routes, and schedules to transport loads of specific size from a location with a pickup time window to a delivery location with a specific delivery time window. The models are customized on a case-by-case basis to reflect a company’s operating environment and customer needs. Recently, dynamic routing and scheduling have grown in importance due to the availability of real-time information from GPS and wireless communication devices. Similar to the network design models described previously, there are few examples of routing and scheduling model implementation among public-sector transportation planning agencies. How- ever, routing and scheduling information at intermodal facil- ities, distribution centers, ports, etc., could greatly improve the estimation of internal freight trips. As shown in Table 2.1, most of these tools are widely used in practice and can be used to answer a number of freight- related planning and policy questions. The exceptions are sup- ply chain/logistics, network design, and routing and schedul- ing models, each of which primarily serves private-sector functions. 2.3 Gaps, Issues, and Challenges Despite the relatively wide use of several model types (time series, behavioral, commodity IO, multimodal network, and microsimulation), the models do not completely meet the 10

needs of public-sector freight planners, modelers, and decision- makers. Key issues include the following: • Lack of a national vision for freight analysis—Since states are conduits for freight movements and regions are impacted by policies and activities originating from outside areas, many DOTs and MPOs stress the need to establish a national vision for freight analysis. Establishment of a national vision for freight demand modeling would help coordinate and guide freight data collection, model con- sistency, and validation/calibration procedures across all public-sector agencies. • Limited ties between freight planning and economic development—There is a need to fully integrate freight demand models with economic models to facilitate trans- portation strategies that maximize a state or region’s eco- nomic advantage. Freight planners, modelers, and decision- makers require quick and reliable methods to determine the economic benefits of transportation investments as well as how economic and accessibility constraints (bottlenecks and employment base) are hindering statewide and regional economic development efforts. • Data limitations—Since freight models often are devel- oped and validated with insufficient data, public-sector agencies and decisionmakers often lack confidence in model results. To improve the statistical validity of their freight models, agencies require more observed data that is gener- ated with greater frequency and accuracy, to conduct more robust model validation. Similarly, agencies require freight data at the appropriate level of detail to support the level of sophistication at which the model is expected to perform. Many agencies that have not yet developed a freight demand 11 Mo de l Cate go ry Description Mode l De ve lo pm en t Mode l Implem en tatio n Public Se ct or Applic at ions Ti me Seri es Shor t- , m ed iu m-, an d lo ng-t er m fore ca sts of fr ei gh t demand an d frei gh t activity Be ha vior al Mode ls ho w co mp an ie s perceive an d se lect from th e ma ny availab le frei gh t ch oi ces. In cl udes ch oi ce - ba se d an d su rv ey -based dema nd mode ls Commodity- Ba se d an d Inpu t-Output Estimat e c urrent an d fore ca sted fr eig ht traf fi c ge nera tio n an d di stri bu tio n by li nk in g in dustri al ac ti vity th rough in put- ou tput mode ls of eco nomi c act ivit y Mu lt im od al Ne twor k Link -nod e n et wo rk represent atio ns of frei gh t suppl y use fu l for determi ni ng tr avel ti mes, costs, reliab ilit y, an d ov er all le vel of serv ic e Mi crosim ul at io n an d Ag en t-Ba se d Mi crosimulation mode ls th e in di vidual movemen t of la rg e number s of un it s an d th eir attr ib utes , wh ile ag en t-based modeli ng defines pote nt ial actors in frei gh t tr an spor at atio n a nd an allowa bl e set of act io ns an d in te ra ctio ns Su ppl y Ch ai n/ Logi stic s Supp ly ch ai ns defi ne the li fe cy cle of products fr om ra w ma terials to th e fina l co ns um er , in clud in g pr oducti on, in ve nt ory , an d tra nsport at io n Ne two rk De si gn Private-sector models for locating factories, distri bu tion ce nt ers, wa re hous es , an d other freight ge nera ti ng fa c ilities Rout in g a nd Sc he du li ng Private-sector models for locating factories, distri bu tion ce nt ers, wa re hous es , an d other freight ge nera ti ng fa c ilities Othe r a nd Emergi ng To pi cs Hybrid models, real-time decision making Widely used, state of the practice Emerging model, limited use Lacking research or application Table 2.1. A comparison of model development and implementation in the literature to public-sector applications.

model, or are considering an upgrade to a more sophisti- cated model indicated that data limitations are a primary obstacle. Specific data needs include – Seasonal trucking variations to account for crop harvest cycles (in rural areas) and consumer demand (in urban areas and around trade gateways); – Time-of-day factors to help evaluate the impacts of policy actions designed to shift truck traffic to off-peak periods or other congestion mitigation strategies; and – Private-sector data to better understand routing and supply chain decisions and impacts of railroads, truck- ing companies, ports, and shippers. • Limitations of existing tools—As described in Table 2.1, existing freight demand modeling and analysis tools are often insufficient to answer freight-related questions being posed by freight planners, freight decisionmakers, and other stakeholders. Critical limitations include – Multimodal network modeling—Agencies need the abil- ity to model multimodal freight flows and interactions, not just light, medium, and heavy trucks. Also needed are dynamic modeling capabilities to evaluate logistics- driven, market-driven, and/or policy-driven mode shifts. Multimodal network modeling would also allow agencies to quantify and compare the burden of each freight mode on the system’s infrastructure. – Behavioral modeling—The conventional four-step travel demand models cannot accurately capture the complex- ities of supply chains and freight systems. They neglect the importance of tour-based and activity-based model- ing. However, few public-sector agencies have developed behavioral models that capture trip chains, less than truckload movements, local truck deliveries, and their associated routings. – Freight routing and route diversion—Existing models are deficient in their ability to assign trucks to the routes they actually use. Similarly, agencies need the ability to esti- mate freight diversion in response to dedicated truck lanes and tolls under different pricing and policy scenarios. – Model adaptability and responsiveness—Freight demand models are too complex, unwieldy, and time-intensive to respond quickly to changing economic conditions as they arise, such as rising fuel costs or facility closures. The time required to develop or update a model is not aligned with the short timeline of freight market demands. There is a need for additional analytical tools that can piggyback on existing models to provide quick-response answers to time-sensitive questions. Similarly, freight models need to be capable of performing various applications and adaptable to the dynamic nature of the freight industry. However, given the complexity of many freight demand models, incorporating new tools or changes into the model is often beyond the capabilities of in-house staff. The subset of people that can actually run the model gets smaller as the model gets more sophisticated. – Temporal variability—Particularly relevant to urban truck models, current freight demand models often lack the ability to evaluate temporal variability, such as time of day and seasonal demand. Regional travel demand models originally developed to support long-range plan- ning did not require time-of-day sensitivities. 12