Freight Demand Modeling: Tools for Public-Sector Decision Making (2008)

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

61 Tour-Based Microsimulation of Urban Commercial Vehicle Movements in Calgary, Alberta, Canada Case Example J. Douglas Hunt, University of Calgary This paper describes the representation of commercial vehicle movements in the Calgary region provided by a tour-based microsimulation system, a working model with a history of use in practical forecasting and policy analysis. The model provides explicit representation of vehicle movements for transport and delivery of both goods and services, with for-hire or carrier services included as the transport sector providing the service of moving goods. The lack of an explicit representation of shipments per se allows some of the complexities associ- ated with such representation to be avoided. Yet the model accounts for truck routes and responds to truck restrictions and related policy. It includes all types of commercial vehicles, from light vehicles to heavier single-unit and multiunit configurations. All sectors of the economy are incorporated into the representation, including retail, industrial, service, and wholesaling. The model has been connected with an aggregate equilibrium model of household-related travel, with the trip tables from the two models assigned jointly to the relevant net- work representations. The microsimulation processes in the model are performed by using external Java applications. The city of Calgary, Alberta, Canada, has a regionaltravel model (RTM) that covers the Calgaryregion. The RTM is used in practical policy analy- sis and forecasting work by both the city and the Alberta Ministry of Transportation. In recognition of the expected benefits, a system for modeling commercial vehicle movements has been developed to work with the RTM and provide representation of the full range of transport of goods and services. This paper presents an overview of the full Calgary RTM with an indication of the position and role of the commercial vehicle move- ment component within the full RTM. A detailed description of the commercial vehicle movement compo- nent using the tour-based microsimulation approach is given, and the calibration of the tour-based microsimu- lation component is discussed. An overview of the result- ing capabilities of the full model is given, and conclusions about what has been achieved and learned in this work with regard to the modeling of movements are presented. CALGARY REGIONAL TRANSPORTATION SYSTEM MODEL The Calgary region is an area centered on the city of Cal- gary and extending approximately 80 km in all direc- tions to include a hinterland of largely agricultural lands dotted with satellite towns and smaller market centers. In 2001 it had a total population of just over 1 million. The Calgary RTM has three basic components: the personal travel demand model, the commercial vehicle movement model (CVM), and the joint vehicle assign- ment process. The personal travel demand model represents the behavior of travelers making trips for household pur- The peer review of this paper was conducted by the Committee on Freight Demand Modeling: A Conference on Tools for Public-Sector Decision Making.

poses, covering about 85 percent of the vehicle trips and vehicle kilometers internal to the Calgary region. The CVM represents the movements of light, medium, and heavy vehicles for commercial purposes, including trans- port and delivery of goods and services, covering about 15 percent of the vehicle trips and vehicle kilometers internal to the region. The personal travel model is an aggregate equilibrium model. It includes representation of 25 travel segments based on person category and movement type. Various private vehicle and transit modes are considered, along with walking and cycling. The CVM, the focus of this paper, is a disaggregate microsimulation model. It includes representation of the tours generated by five categories of industrial activity on each of five types of land use. The individual trips on each separate vehicle tour are simulated, providing a vehicle type, an origin, a destination, and a time of trip, among other attributes, for each such trip. The joint vehicle assignment process loads the trip tables generated by the above two demand models to a nodes-and-links representation of the road networks in the region, establishing a network equilibrium loading taking account of the congestion on links. Five time peri- ods are considered in the assignment process, including the busiest 1⁄2 hour (the “crown”) and the rest of the 11⁄2 hours (the “shoulders”) for the a.m. peak period (from 7 to 9 a.m.), the similar busiest 1⁄2 hour and the rest of the 11⁄2 hours for the p.m. peak period (from 4 to 6 p.m.), and the off-peak period covering the rest of the day. The congested travel times from the network for each of these time periods are fed back into these models, and the process is iterated until the travel times used by the mod- els are consistent with those arising from the subsequent loading on the networks, thereby establishing a system equilibrium. Within each iteration where the congested travel times are fed back, the personal travel model is run once to equilibrium and the resulting trip table output while the commercial vehicle model microsimulation is run 10 times and the results are averaged to obtain expected val- ues for the zone-to-zone trips in the trip tables. In the final iteration of the full modeling system, the microsim- ulation is run 30 times and the results are averaged to obtain expected values with better statistical properties. There is a fourth component dealing with vehicle trips with at least one end external to the Calgary region. It is a fairly modest set of singly constrained gravity models considering the exogenously forecast vehicle flows pass- ing through the model external cordon entry and exit points, which account for about 6 percent of the total vehicle trips in the entire region. This generates addi- tional vehicle trip tables for each of the light, medium, and heavy vehicle categories for each time period in assignment, and these trip tables are combined with those from the personal and commercial models before the assignment performed in each iteration. STRUCTURE OF THE MICROSIMULATION OF COMMERCIAL VEHICLE MOVEMENTS The microsimulation of commercial vehicle movements in the CVM considers the tour-based movements by using Monte Carlo techniques to assign the attributes to each tour in a list of tours generated for each zone, including tour purpose, vehicle type, next stop purpose, next stop location, and next stop duration. Overall Structure The overall framework of the microsimulation process is illustrated in Figure 1. First, the number of tours based in each zone is estab- lished by using an aggregate trip generation model. This value establishes the length of the list of tours whose spe- cific attributes are identified one after another as the microsimulation progresses. Then, on the basis of Monte Carlo processes, each tour in the list for each zone is con- sidered, one at a time, and the vehicle type and purpose of the tour are identified, followed by the specific tour start time. The characteristics of the stops on the tour are then identified, iterating stop by stop until the tour is finished. Tours are “grown” incrementally by having a ”return- to-establishment” alternative within the next stop pur- pose allocation: if the next stop purpose is not return to establishment, then the tour extends by one more stop. 62 FREIGHT DEMAND MODELING: TOOLS FOR PUBLIC-SECTOR DECISION MAKING FIGURE 1 Overall tour-based microsimulation framework. Stop Duration Iterates to “grow” tourNext Stop Location Next Stop Purpose Tour Start Vehicle and Tour Purpose Tour Generation

This approach is more consistent with the nature of tour making in urban commercial movements—where there are a comparatively large number of equally important stops in many tours. This is in contrast to the ”rubber- banding” process typically used with household tour- based modeling, where first a primary destination for the tour is established out from the base and then perhaps one or two intermediate stops on the trips between the base and this primary destination are identified— analogous to first stretching a rubber band between two points and then pulling it wider along the lengths in between (1). The selection probabilities used in the microsimulation processes are established on the basis of logit models estimated by using the choice data collected in the surveys for different segments of the full range of commercial movements. Time is treated as a continuum, rather than in discrete periods, and both start and end times are established for each trip and each stop on each tour. Development Data The primary source of the data used in development is an extensive set of interviews about own-account commer- cial vehicle movements conducted at just over 3,100 busi- ness enterprises in the Calgary region—analogous to household trip diary interviews—that collected informa- tion on tours made on a typical weekday in 2001 (2). Sam- pled establishments provided information on the movements of their entire fleet over a 24-hour period, including origin, destination, purpose, fleet, and commod- ity information. The resulting sample provided choice behavior information on just over 64,000 commercial vehicle trips for use in the estimation and further calibra- tion of the model components. The data were expanded by industry, size, and location to represent the total popu- lation of commercial enterprises, which was challenging in itself because of the uncertainty surrounding the total pop- ulation of employment at establishments (3). Terms, Categories, and Basic Values Three categories of vehicle are considered: • Light vehicles: small four-tire vehicles (cars, vans, pickups, and SUVs); • Medium vehicles: single-unit trucks with six tires; and • Heavy vehicles: multiunit trucks with more than six tires. Four stop and related trip purposes are considered— in much the same way that work and school purposes are considered in personal travel modeling: • Goods: goods delivery or pickup, including goods- handling and transport activities; • Services: service delivery, including any incidental materials handling (such as an electrician picking up elec- trical supplies); • Others: all nondirect goods and services activities not included in the above or at the point where the tour started, including breaks, meals, vehicle fueling, and so forth; and • Return to establishment: returning to the starting point of the tour, either at the end of the day or during the day, for any reason. These different commercial movement purposes relate to different types and distributions of activities, which imply different types of companies with different options, objectives, influences, and choice structures. The business establishments and the associated employment at these establishments are segregated into five establishment categories on the basis of the two-digit sector-level categories in the North American Industry Classification System (4) as follows: industrial, whole- sale, retail, transport, and services. Each of these five categories of establishment is handled separately throughout the microsimulation, each with a largely unique set of coefficients throughout the process, so the results are different, with different behaviors and reac- tions to policy changes, for these categories. The frame- work itself is also slightly different for the transport category in particular. The transport category includes what are called “for-hire” or “private” carriers, in essence trucking companies that sell transportation service. These are different in that the goods and services stop and tour purpose categories are combined into a single “business” purpose—in recognition of the fact that transport estab- lishments provide the service of handling goods, which blurs the definitions. The zones in the model are classified into five land use types on the basis of specific zonal attributes as follows: • Low density, • Residential, • Retail and commercial, • Industrial, and • Employment node. These land use types are used to differentiate coefficient values and resulting model sensitivities at various points in the microsimulation. They work in combination with the establishment categories to separate the blue-collar and white-collar components of given industries, which allows the microsimulation to differentiate between the patterns of commercial movements arising from these components. Travel utilities that are weighted combinations of travel times and travel distances are used throughout in 63TOUR-BASED MICROSIMULATION IN CALGARY

the representation of travel conditions for movements between zones. The weights used vary by vehicle type and are always negative, consistent with travel having a gen- eral cost. In this sense the travel utility for a trip as used here is the negative of the generalized cost of the trip. Vehicles in the medium and heavy categories are sub- ject to truck route restrictions on the road network in Calgary. Drivers of these vehicles must minimize the dis- tance they travel on the portions of the road network that are not designated truck routes. For links that are not des- ignated truck routes, a large fixed penalty is added to the generalized cost faced by medium and heavy trucks for each additional 50 meters of the link used, so that the net- work assignment process respects these restrictions. The penalty portions of the resulting travel times are then removed from the network skims so that representations of the actual times and distances are used in the rest of the microsimulation process. Tour Generation In this first step, the aggregate number of tours generated by each category of establishment is determined for each time period in each model zone. These numbers are used to form lists of discrete tours considered in the rest of the model. The tour generation rate (tours per employee) is first determined for the entire day for each category of estab- lishment for each zone by using an exponential regression equation with zonal attributes as the independent vari- ables. This rate is multiplied by the number of employees in the relevant category of establishment in the zone to produce a total number of tours generated in the zone by that industry for the entire day. The attributes represented in the exponential regression equations include the land use type for the origin zone, the percentage of zonal employment in the same establishment category in the ori- gin zone, and accessibilities to total employment for the origin zone. These numbers of tours for the entire day are then split among time periods covering the day to establish the number of tours in each time period by each category of establishment in each zone. The time periods consid- ered are early off peak, midnight to 7 a.m.; a.m. peak, 7 to 9 a.m.; midday off peak, 9 a.m. to to 4 p.m.; p.m. peak, 4 to 6 p.m.; and late off peak, 6 p.m. to midnight. The splits among time periods are determined by using logit models, with utility functions that include the same sorts of zonal-level attributes used in the exponen- tial regression equations indicated above. In each case the resulting number of tours in each time period by each category of establishment in each zone becomes the length of the list of corresponding discrete tours of that type, whose remaining attributes are estab- lished in the rest of the microsimulation process. Tour Purpose and Vehicle Type Allocation In this second step, each tour in the lists for each zone is assigned both a primary purpose and a vehicle type. A Monte Carlo process is used to assign both simultane- ously, where the selection probabilities are determined by using single-level logit models based on establishment category with utility functions that include zonal-level land use, establishment location, and accessibility attributes. The alternatives for the primary purpose for a tour are • Goods, • Service, • Other, and • Fleet allocator. The first three of these categories are consistent with the stop purpose definitions indicated above. The last, fleet allocator, includes tours by vehicles where the data collection process sought indications of more general vehicle use statistics rather than each stop and the travel to and from it, in recognition of the large collection bur- den that would be imposed, as in the case, for example, of newspaper delivery, postal services, and refuse collection. The alternatives for the vehicle type for a tour, again consistent with the vehicle category definitions indicated above, are light, medium, and heavy. Tour Start Time As described above, in tour generation, lists of tours are allocated to one of five time periods. In this step, each tour in the list for each time period is assigned a precise start time. This is done by using a Monte Carlo process with sampling distributions based on the weighted sam- ple of observed start times differentiated by establish- ment category and time period. A cumulative percentage distribution function was calculated by industry and time period on the basis of a curve fit to observed data. These sampling distributions are static, which implies that changes in the temporal distribution for the starts of tours established by the microsimulation in response to changes in travel conditions (or any other potential policy options for that matter) are limited to the changes in the time period allocations in trip generation. But there is fur- ther potential for travel conditions to influence the times for the rest of a given tour. The microsimulation keeps track of the precise times for the arrival and departure at each subsequent stop on each tour. This includes using the travel time between each stop. To the extent that travel times on the network change in response to policy 64 FREIGHT DEMAND MODELING: TOOLS FOR PUBLIC-SECTOR DECISION MAKING

inputs (or any other influences), the arrival times at sub- sequent stops will also change, which can lead to changes in the decision made with regard to the next stop purpose as described below. Further, as tours continue, the microsimulation will allow them to cross into the next time period. For example, a vehicle can start a tour in the a.m. peak and then eventually find itself in the midday off peak, where improved travel conditions can further affect the purposes and locations of subsequent stops. After the tour start time has been assigned to a given tour, the microsimulation begins the iterative process of “growing” the tour by assigning sets of next stop pur- pose, next stop location, and next stop duration until the next stop purpose is “return to establishment.” Next Stop Purpose The purpose for each subsequent stop is assigned from the following alternatives, with restrictions on availabil- ity as indicated: • Goods: available if the primary purpose of the tour is goods; • Service: available if the primary purpose of the tour is service; • Other: available if the primary purpose of the tour is goods, service, or other; and • Return to establishment: if the next stop is not the first stop on the tour. The term “business stop” is used here to refer to stops that are either goods stops (when the tour primary pur- pose is goods) or service stops (when the tour primary purpose is service). Again, a Monte Carlo process is used to assign the next stop purpose, with the selection probabilities determined by using single-level logit models based on a “segment” category. With so many observations of next stop pur- pose available, it was possible to estimate utility function coefficients for 13 segments of commercial movements based on combinations of industry category, vehicle type, and tour primary purpose, consistent with differences in the influences on next stop choice behavior, as follows: • S-S-L: service tours by services establishments using light vehicles; • S-S-MH: service tours by services establishments using medium or heavy vehicles; • G-S-LMH: goods tours by services establishments using any vehicle type; • S-R-LMH: service tours by retail establishments using any vehicle type; • G-R-LMH: goods tours by retail establishments using any vehicle type; • S-I-L: service tours by industrial establishments using light vehicles; • S-I-MH: service tours by industrial establishments using medium or heavy vehicles; • G-I-LMH: goods tours by industrial establishments using any vehicle type; • S-W-LMH: service tours by wholesale establish- ments using any vehicle type; • G-W-L: goods tours by wholesale establishments using light vehicles; • G-W-MH: goods tours by wholesale establish- ments using medium or heavy vehicles; • B-T-LMH: business tours by transport establish- ments using any vehicle type; and • O-X-LMH: other tours by any establishments using any vehicle type, including fleet allocator tours. The utility functions for the next stop purpose alter- natives in the logit models include representation of the following attributes: • Number stops for business purposes made previ- ously in the tour; • Number of stops for other purposes made previ- ously in the tour; • Number of stops for any purposes made previously in the tour; • Elapsed total time for the tour to that point, which is the total time that has been spent on the tour up to that point, including all time spent at stops and in travel between stops up to that point; • Elapsed travel time for the tour to that point, which is the total time that has been spent traveling on the tour up to that point, including all time spent in travel between stops but not including all time spent at stops up to that point; • Travel utility associated with making the trip from the current location zone to the zone where the tour began for the vehicle type being used; and • Accessibility for the current location (zone) to all categories of employment in all zones for the vehicle type being used. Next Stop Location After the next stop purpose has been assigned, the next stop location is assigned—if the next stop purpose is not return to establishment. The available alternatives for the next stop location are the 1,447 model zones. Again, a Monte Carlo process is used, with the selec- tion probabilities determined by using single-level logit models based on 13 “segment” categories similar to those used in the selection of next stop purpose. In this case the 13 segment categories are based on combinations of indus- 65TOUR-BASED MICROSIMULATION IN CALGARY

try category, vehicle type, and next stop purpose (not tour primary purpose), with the goods, service, and “other” categories still being used, but in this case for the assigned next stop purpose (rather than the assigned tour primary purpose). The 13 category definitions remain the same— apart from using stop purpose rather than tour primary purpose—so the designations for the categories still apply: thus, for example, the S-I-L category in this case indicates “service stops made on tours by industrial establishments using light vehicles” (whereas previously, in the case of next stop purpose, it indicates “service tours made by industrial establishments using light vehicles”). With these 13 segments, different logit models are used for the assign- ment of next stop location depending on whether the next stop purpose is goods, service, or “other,” thereby allow- ing the appropriate spatial distribution of opportunities to be taken into account, even on the same tour. The utility functions for the next stop location (zone) alternatives in the logit models include representation of the following attributes: • Land use type for the possible next zone; • Accessibility to all categories of population for the possible next zone for the vehicle type being used; • Accessibility to all categories of employment for the possible next zone for the vehicle type being used; • A numerical score representing the relative attrac- tiveness of the possible next zone for stops made during tours generated by transport establishments, which is determined as described further below; and • The “enclosed angle” for the possible next zone, which is the angle (in degrees) enclosed by (a) the straight line from the current zone to the zone containing the establishment and (b) the straight line from the current zone to the possible next zone (an example of this angle is shown in Figure 2); a value of 0° indicates that the pos- sible next zone is in the same direction as the zone con- taining the establishment, and a value of 180° indicates that the possible next zone is in the opposite direction from the zone containing the establishment. Stop Duration Model In this step, the stop being considered is assigned a pre- cise duration. This is done by using a Monte Carlo process with sampling distributions based on the weighted sample of observed durations differentiated by the 13 segments also considered in the assignment of next stop purpose and next stop location. The microsimulation uses the precise duration assigned to the stop to advance the clock keeping track of start and end times and then begins another iteration for the next stop. CALIBRATION OF THE MICROSIMULATION OF COMMERCIAL VEHICLE MOVEMENTS After all the elements of the microsimulation process were assembled and the values for the various coeffi- cients established, the entire process was calibrated to match various aggregate targets appropriately. An iterative approach was used under which the process was run, the match of the output values to spe- cific aggregate targets assessed, and the associated category-specific constants adjusted to improve the match. With Monte Carlo processes like the one described here, in general the results are different with each run. Therefore, multiple runs were done and the results averaged to get values that indicate the central tendencies of the outputs. Initial experimentation showed that in this case averaging over 10 runs provided highly stable results, with variations on the order of 1 percent related to the aggregate targets being considered. The elements of the microsimulation are interdepen- dent, which means that adjustments to the values of the coefficients in one element can alter the output values for other elements. For example, if the tour generation is adjusted, establishment locations are changed, which affects the decision to return to establishment and there- fore tour lengths. This led to the use of an approach in calibration under which the matches to different sets of targets were considered consecutively over a series of iterations until the adjustments to coefficients and the resulting changes in output values were small enough to be of no consequence. The following sets of aggregate targets were considered in the order indicated: • Tour generation by industry and geographic area; • Proportions of tours starting in the a.m., p.m., and combined off-peak periods; • Vehicle type and tour purpose proportions; • Number of stops per tour by 13 segments; • Total trip destinations in each of 13 superzones by vehicle type (for example, the proportion of all trips by heavy vehicles that are destined to the southeast indus- trial area); • Intrasuperzonal proportions of trips within each of the 13 superzones by vehicle type (for example, the pro- portion of light vehicle trips with destinations within the 66 FREIGHT DEMAND MODELING: TOOLS FOR PUBLIC-SECTOR DECISION MAKING current zone possible next zone zone containing establishment enclosed angle FIGURE 2 Example of enclosed angle for possible next zone.

central business district that also originated within the central business district); and • Total trips by vehicle type and industry. The matches to observed aggregate values were within reasonable margins in all cases and within a fraction of a percent in a large majority of cases. Figure 3 shows the results of the calibration with regard to the intrasuper- zonal proportions of trips for the 13 superzones. Figure 4 shows the results after calibration with regard to the number of stops for the 13 segments. Figure 5 shows the changes in match for tour purpose and vehicle type pro- portions by employment category as the iterations in cal- ibration proceeded. 67TOUR-BASED MICROSIMULATION IN CALGARY 0 10 20 30 40 50 60 70 80 90 100 CB D CB D F rin ge NE -In du st. Ce nt- Ind ust . SE -In du st. NW - Re sid . N- Re sid . NE -R es id. S-R es id. SE -Re sid . W- Re sid . Re gio n-N Re gio n-S Superzone In tr as up er zo na l % o f t rip s Before After Target 0 10,000 20,000 30,000 40,000 50,000 60,000 O- X-L MH S-S -L S-S -M H G- S-L MH S-R -LM H G- R-L MH S-I -L S-I -M H G- I-L MH S-W - LM H G- W- L G- W- MH B-T - LM H Model Sector To ta l T rip s Before After Observed 87,500 FIGURE 3 Match of tour-based microsimulation results to intrasuperzonal proportion of trip (medium vehicle) targets at start and end of calibration (CBD = central business district). FIGURE 4 Match of tour-based microsimulation results to number of stops by segment at start.

Figure 6 shows the link-level flows of heavy vehicles only obtained when the full Calgary RTM is run, includ- ing the CVM, the personal travel model, and external- internal component, for the model base year. This resulting assignment provides a good fit with observed patterns—closely matching observed flows and display- ing a focus on industrial areas (those with darker shade) and an adherence to truck routes. MODEL CAPABILITIES The calibrated tour-based microsimulation process for the commercial vehicle movement component, together with the other calibrated components of the Calgary RTM, provide a representation of the transportation sys- tem in the Calgary region that can be used in both fore- casting and policy analysis. Its application in forecasting requires inputs concerning population, employment, and transport supply conditions similar to those required for the forecasting of household travel alone, along with information concerning truck route policy and vehicle- specific values of time and distance-based operating costs for commercial components. For the analysis of policy affecting commercial move- ments, this representation will respond to changes with regard to • Road network capacities and connectivity; • Truck route policy; 68 FREIGHT DEMAND MODELING: TOOLS FOR PUBLIC-SECTOR DECISION MAKING 0 5 10 15 20 25 0 1 2 3 4 5 Iteration # Av er ag e o f a bs ol ut e % v ar ia nc e, al l p ro po rti on s Service Industrial Retail Transport Wholesale Overall FIGURE 5 Changes in match of tour-based microsimulation results to proportions of tours by tour purpose and vehicle type targets over early series of iterations in calibration. FIGURE 6 Plot of heavy-vehicle flows resulting from full Calgary CVM. Simulated patterns closely match observed flows and display a focus on industrial areas (shown in darker shade) and an adherence to truck routes.

• Road tolls; • Fuel taxes; • Household travel (resulting in changes in roadway congestion); • Population level and spatial distribution; and • Employment level, composition, and spatial distribution. The responses to such changes will occur in multiple elements of the microsimulation. Tour generation, the allocation to start time period, tour purpose and vehi- cle type choice, next stop purpose, and next stop loca- tion all respond to changes in travel conditions. Thus, if travel conditions become more onerous for commer- cial movements—perhaps because the network becomes more congested or because a key part of the truck route system is removed—commercial vehicles will not merely travel shorter distances; they will also make fewer stops per tour and more tours to fulfill the demand. CONCLUSIONS The CVM described here demonstrates the practical fea- sibility of using a tour-based microsimulation approach in the modeling of commercial vehicle movements in a novel way that allows the incorporation of representa- tions of these influences. The following are some of the notable aspects of the model: • A tour generation element that includes a response to changes in transport conditions such that (in the short run) more tours arise when travel times increase; • Variation in tour primary purpose and vehicle choice across a broad spectrum of activities and in response to changes in employment, population, and resulting accessibilities; • A “growing” of tours more consistent with the nature of commercial movements with potentially larger numbers of equally important stops, as opposed to the ”rubber-banding” process typically used in the represen- tation of tours of household movements; • Representation of the influence of tour duration and, at least partially, the time of day on tour patterns; • Consideration of the physical shape of tours; • Responsiveness to changes in truck route policy as well as infrastructure and cost changes specific to three categories of commercial vehicle; • Separation of the fleet allocator and shipment- focused components of commercial movements; • A range of interactions among the elements such that changes to the inputs affect the simulated behavior in a variety of dimensions; and • A set of alternative specific constants for each ele- ment that allows calibration of the full microsimulation system to aggregate targets. One of the advantages of this modeling approach is that it does not rely on any explicit representation of shipments or related transactions. Dealing with ship- ments, translating from commodity flows to shipment sizes to vehicle allocations, introduces a number of com- plexities. Some impressive work has been done by others seeking to represent these complexities. The approach used here bypasses much of the need for this additional complexity by focusing on vehicles through the use of generation rates and vehicle allocation models that implicitly take much of this into account parsimoniously. A complete and accurate representation of the full range of factors influencing the translations from commodity flows to shipment sizes to vehicle allocations would pro- vide a model with a more robust policy responsiveness, but in a practical setting, the model described here is in many cases a more realistic solution. At this point the system is being used in practical pol- icy analysis work. The expectation is that more will be learned about the capabilities of the model and its use as this work progresses and that the need for further improvements will be identified. In addition, the success- ful implementation of this model in Calgary suggests the potential for successful implementation elsewhere—in fact, models based on this approach and structure are under development for Edmonton in Canada and Ohio in the United States. This has included reusing (with suit- able recalibration) the destination choice components in particular—where the greatest amount of data manipu- lation and work arises. Even without further improvements, the current sys- tem provides a useful tool, taking both the representa- tion and the associated understanding of urban commercial movements well beyond the freight-only, large-truck-only, and regional-level approaches used pre- viously. It permits a much richer treatment of relevant aspects such as the importance of trip chaining and less- than-load hauling, the significance of service delivery as a motivator for travel, and the role of light commercial vehicles. The system points a way ahead in the modeling of the commercial vehicle sector of the urban trans- portation system. ACKNOWLEDGMENTS The City of Calgary, the City of Edmonton, and Alberta Transportation funded much of the development work described here, along with the associated data collection. Kevin Stefan of the City of Calgary did much of the work in developing the microsimulation system described. 69TOUR-BASED MICROSIMULATION IN CALGARY

Dianne Atkins, Ali Farhan, David Kim, Douglas Mor- gan, and Paul McMillan of the City of Calgary and John Abraham of the University of Calgary also contributed to the work. Some of the research and the preparation of this paper were funded from grants provided by the Natural Sci- ences and Engineering Council of Canada and the Insti- tute for Advanced Policy Research at the University of Calgary. The material presented in this paper in places draws heavily from certain components of a manuscript sub- mitted to the Transportation Research Record: Journal of the Transportation Research Board for consideration for publication. REFERENCES 1. Jonnalagadda, N., J. Freedman, W. A. Davidson, and J. D. Hunt. Development of Microsimulation Activity- Based Model for San Francisco: Destination and Mode Choice Models. In Transportation Research Record: Journal of the Transportation Research Board, No. 1777, TRB, National Research Council, Washington, D.C., 2001, pp. 25–35. 2. Hunt, J. D., and D. G. Morgan. The Calgary Region Commodity Flow Survey. Proc., 2001 Annual Conference of the Canadian Institute of Transportation Engineers, Calgary, Alberta, Canada, May 2001. 3. Hunt, J. D., K. J. Stefan, and A. T. Brownlee. Establishment-Based Survey of Urban Commercial Vehicle Movements in Alberta, Canada: Survey Design, Implementation, and Results. In Transportation Research Record: Journal of the Transportation Research Board, No. 1957, Transportation Research Board of the National Academies, Washington, D.C., 2006, pp. 75–83. 4. North American Industry Classification System— Canada 1997. Statistics Canada, Ottawa, Ontario, Canada, 1998. DISCUSSION Scott Drumm, Port of Portland This discussion is framed by a view of the Calgary regional travel model from a freight data perspective and from professional experience working with a truck model in Portland, Oregon. Thus, the comments are greatly influenced by both perspectives. The review is organized on the basis of five key themes important for freight modeling: • Greater coverage of truck types and activities, • Linkage between freight and land use, • Connectivity with economic models, • Ease of data acquisition, and • Evolution from truck to freight models. The first theme concerns having greater coverage of truck types, sizes, and activities in freight models, including • Smaller trucks, • Interactions with heavy-truck trips, • Truck destinations, and • Coverage of nonfreight stops (services). At the local level, policy makers as well as business leaders are becoming more interested in understanding the movement of not only heavy trucks but also com- mercial and service trucks. These vehicles generate the bulk of truck trips on a local or regional transportation system. Understanding their role, behavior, and needs will be necessary if freight modeling is to help build an accurate picture of goods movement. The Calgary model leads us in that direction with its inclusion of a broad range of truck types and nonfreight (service) stops. This element of intraregional trips and small truck move- ments is an area not well covered in most models. Linkage between freight and land use is becoming increasingly important in public policy and urban plan- ning discussions. Land use is a prime factor in determin- ing where freight moves. As land at key freeway interchanges, near inter- and multimodal facilities, and in zoned industrial areas becomes more scarce, freight models linked to land use have an important role in help- ing regions and localities determine the trade-offs between various transportation and land use decisions. The Calgary model, as well as the Oregon statewide model (see Oregon Generation 1 Land Use–Transport Economic Model Treatment of Commercial Movements: Case Example), has begun this convergence. One area where freight models struggle is with con- nectivity with economic models. The structure and per- formance of the model area’s economy influence traffic volumes generally, modal volumes specifically, and trip geography. Furthermore, the economy and economic models serve as the basis for projections of freight growth. If one is trying to sort out what is happening on the surface transportation system at the present, this tie is not necessary. On the other hand, if a model is to be used to project where trucks are going to be in the future, understanding where the economy is headed is essential. The Calgary model lacks this link, but its use is not as much oriented toward the future as it is toward under- standing how the system functions today. The Oregon statewide model, however, is trying to project the future, and it links to economic inputs as its first step. 70 FREIGHT DEMAND MODELING: TOOLS FOR PUBLIC-SECTOR DECISION MAKING

An important consideration in the development of a model is the ease of data acquisition. If a model is to be kept current, the data supporting it must be readily avail- able and easy to obtain. If this is not the case, updating the model will become costly. The model structure must also be able to accommodate data updates easily. How- ever, there is a trade-off. If one builds a model such that inputting new or updated data is easy and acquiring that data is simple, one likely sacrifices accuracy. This is shown in the Calgary model. The data are detailed and were time-intensive to collect, thus making future updates more challenging. The accuracy, though, bene- fits from this investment in time and resources. The Ore- gon model is based on commodity flow forecasts, which are relatively easy to obtain, but because it does not have the level of detail or reliance on primary data, its accu- racy is somewhat lessened. If models are to be truly effective in helping make investment decisions with regard to goods movement, evolution from truck to freight models must occur. Most models, such as Calgary’s and the Portland truck model, focus on trucks. Although there are many challenges in developing models such as these, the practice will need to move toward multimodal freight models. This will afford the ability to understand and predict mode shift, determine where investments in nonroad modes will ben- efit the road and highway system, and understand how changes in one mode or its facilities cascade throughout the goods movement system. In conclusion, different models serve different pur- poses, and there are many levels of sophistication in goods movement models. The ultimate objective is to answer the right questions with the right models. Under- standing what question is to be answered becomes the primary factor in determining how to build a model or which model to use. Every model has its strengths and weaknesses, and any given model may not meet a spe- cific area’s needs. On the basis of an understanding of the questions that the Calgary region needed to answer, the model yielded the kind of outputs that were sought. It also provides ideas that can be applied in other areas. Through its accounting for a variety of commercial vehi- cle types and activities, the commercial vehicle compo- nent of the Calgary regional travel model is moving in the right direction. The views expressed in this paper are entirely those of the author and do not necessarily indicate the positions of any of the sponsoring agencies. Any errors or omis- sions are also solely the responsibility of the author. 71TOUR-BASED MICROSIMULATION IN CALGARY