An Urban Commercial Vehicle Movement Model for Calgary KJ Stefan City of Calgary PO Box 2100, Stn. M, #8124 Calgary, Alberta T2P 2M5 Phone: 403-268-1630 Email: firstname.lastname@example.org JDP McMillan City of Calgary PO Box 2100, Stn. M, #8124 Calgary, Alberta T2P 2M5 Phone: 403-268-5985 Email: email@example.com JD Hunt University of Calgary 2500 University Drive NW Calgary, Alberta T2N 1N4 Phone: 403-220-8793 Email: firstname.lastname@example.org Paper for presentation at the 84th Transportation Research Board Annual Conference Washington DC, January 2005 ABSTRACT Commercial vehicle movements comprise perhaps 15% of all urban vehicle trips, and produce even larger impacts in key areas such as congestion, emissions, road wear and industrial area traffic. A system for modelling commercial movements has been developed for Calgary in Canada. The resulting system of models includes the novel application of an agent-based microsimulation framework, using a tour-based approach and emphasizing important elements of urban commercial movement, including the role of service delivery, light commercial vehicles and trip chaining. The microsimulation uses Monte Carlo techniques to assign tour purpose, vehicle type, next stop purpose, next stop location, and next stop duration. Tours are ‘grown’ using a return-to-establishment alternative within the next stop purpose allocation, which is seen to be consistent with the nature of tour making in urban commercial movements. The Monte Carlo probabilities are established using a series of logit models, with coefficients estimated based on the observed behavior of different segments of commercial movements. The estimation results in themselves provide insights into the revealed behavior that have not been available previously. 1 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal. INTRODUCTION Commercial vehicle movements are a significant portion of travel in urban areas. It is estimated that about 10 to 15 percent of urban vehicle trips are made for commercial purposes. But the impacts of commercial vehicles are even greater than this: commercial vehicles (and especially the larger trucks) tend to concentrate in industrial and commercial areas, and more during the middle of the workday than in the peaks. Additionally, when compared with the personal vehicle fleet, larger trucks can have more significant impacts per vehicle in key areas such as road congestion and traffic flow, greenhouse and pollutant emissions and pavement wear. Higher values of time are also typically attributed to commercial vehicles, so this sector should be considered separately in order to properly calculate the benefits of travel time savings. This paper uses ‘commercial’ in an inclusive sense, to mean not only trips made by commercial enterprises, but also by non-commercial organizations such as governments and government agencies, as well as trips by employees of charities and similar non-commercial organizations. A trip is considered to be ‘commercial’ if the trip maker is being reimbursed for making the trip above and beyond the reimbursement of travel costs. These trips are considered separate from the ‘personal’ trips produced by other elements of the overall Calgary model system. The Calgary census metropolitan region has a population of around one million. Calgary is a key hub for shipping in Western Canada, with key strategic highway and railway routes for commercial traffic. The City of Calgary has developed a regional travel model, focussing on personal travel patterns on a typical fall weekday. However, the treatment of commercial vehicle movements prior to the development of the Calgary Commercial Vehicle Model (CVM) was very limited, using a scaling of truck flows derived from count data. A limited treatment of commercial movements is typical of virtually all models of this sort, despite the importance of these movements as indicated above. MODEL DESIGN Urban Commercial Vehicle Movement Commercial vehicle movement is very different from personal movement in many respects. Urban commercial movement is also different from commercial movement over larger areas, such as states, provinces, nations and international areas. Rail, ship and aircraft transport are inefficient over the smaller scale of urban areas, hence urban commercial movement is almost exclusively road-based (1). The bulk of the work in the field of commercial movement (or freight) modelling is done on a large regional scale, but importing these approaches to the urban context is inappropriate given the differences that exist. A large component of commercial movements within urban areas is made with light commercial vehicles (LCVs), including four-tire, two-axle vehicles such as pick-up trucks and vans and even passenger cars. Calgary data indicates that over 50% of urban commercial trips are made by light vehicles. Interurban transport makes load consolidation more economic, and the larger volumes of goods moving between larger areas (2) also increases the importance of heavy commercial vehicles such as tractor-trailer combinations over longer distances. Further, the service sector constitutes a much larger large proportion of the urban economy, in large part serving the population in the urban area where it is located. While almost all interurban transport is goods hauling, Calgary surveys revealed that approximately 50% of all business stops were made to provide a service. It is clear that a complete consideration of urban commercial movement requires consideration beyond just freight movements, expanding to include service deliveries with the urban area. Methods of Modeling Urban Commercial Vehicle Movement The most common methods for urban commercial vehicle modeling are various forms of expanded OD- matrix techniques, typically based on ground counts expanded to represent a steady-state and perhaps further scaled for future years (3). This general approach has several problems; the most obvious are the lack of policy response and the exclusion of the half of commercial movements that are made with light commercial vehicles. These models 2 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal. are perhaps only useful for providing ‘background traffic’, where personal vehicle modeling is the goal and commercial vehicles are of little interest beyond their consumption of roadway capacity. The second major methods use variations on the traditional four-step aggregate approach (4). Mode choice is often neglected entirely; usually only larger vehicles are considered. This approach emphasizes trips rather than tours, which misses the essential tour-based nature of urban commercial vehicle movements. Additionally, most models based on this approach tend to use simple ‘standard’ coefficients to handle tour generation and trip distribution, and often neglect service trips, focussing on goods movement alone. While most jurisdictions use one of the above-described modelling approaches for commercial vehicles (or neglect them entirely), a number of more novel approaches are being developed. The approach in supply chain modelling is to attempt representation of the suppliers, warehouses and consumers of products in individual supply chains (5). This would require substantial, hard-to-get data to produce a comprehensive model for a typically diverse North American urban area. Some models use the spatially disaggregate input-output approach, which does a good job of representing the commodity flows in the economy, but typically downplays or simplifies important elements of urban commercial movements, such as trip chaining and less than load hauling (6). The microsimulation approach has recently become the state-of-the-art in practice in household travel modelling (7). It is more flexible and powerful, but in general requires a richer dataset in order for these benefits to be realized effectively. Some of the related advantages of a microsimulation approach are the ability to model various aspects of choice behaviour explicitly, aggregate results in any manner desired, and the flexibility to create any number of specific constraints. The potential for a tour-based microsimulation model of urban truck / freight movements has been discussed in the literature (8) and some of the possible components of such a model partially explored empirically (9) but there do not appear to be any successful, practical implementations up to now. Calgary Commercial Vehicle Movement Model Design For the Calgary commercial vehicle movement model, a hybrid approach was developed with a tour-based microsimulation included. The total set of commercial vehicle movements was divided into three separately modelled groups: tour-based movements, fleet-allocator movements and external-internal movements. The majority of movements are covered by the tour-based model, with the two other movements being modelled with methods that use the available data. The external-internal movements model uses singly-constrained gravity models to assign the internal ends of trips being made by trucks crossing the model area boundary. This model only includes trucks – and excludes those commercial trips with at least one end external by light vehicles – because the external cordon survey done to collect the information used to develop this model only considered trucks; light vehicles were not stopped. These external movements comprise roughly 6% of all movements. Fleet allocators represent roughly one quarter of all commercial vehicle movements. Fleet allocators primarily dispatch vehicles to cover an area or to travel road links, rather than handle a specific shipment. Examples include newspaper delivery, garbage pickup, mail and courier service, taxi, police and rental car fleets. At this point the fleet allocator model uses an aggregate generation and gravity-style distribution framework. The remaining two thirds of trips are modelled using a tour-based microsimulation. This model identifies the attributes of individual tours, including stop purpose, vehicle type, time of day and the stop locations. It uses a Monte Carlo simulation with a series of logit models to simulate the decisions made with a touring vehicle. Because of the microsimulation element, results can be aggregated in any way to analyse tours for different industries, time periods or other results as needed. The tour-based microsimulator is the focus of this paper. Figure 1 illustrates the overall structure of the model. Tours are generated on a zonal level, then each tour is assigned a vehicle and overall tour purpose. A start time is then determined. The trips in the tour are then identified in an iterative process. For a given tour, only one vehicle type and one primary purpose are selected, but at least one and possibly more stop locations and purposes are selected, with the previous and current stops influencing the decisions on the next stop. To implement this, an iterative loop is formed with the next stop purpose being chosen, then the next stop location, then the duration of the stop at the destination. 3 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal. Tours are ‘grown’ in this implementation, rather than developed using the ‘rubber-banding’ approach commonly seen in personal travel modelling – where a primary destination is identified first and one or more further intermediate stops are identified out from the paths between the tour base and the primary destination after that (7), like stretching a rubber band first between the tour base and the primary destination and then possibly outwards on either side. This reflects the nature of commercial vehicle movements – rather than having a work or other compulsory purpose to anchor one end of the tour and perhaps one or two optional intervening stops, a range of equally important and possibly compulsory stops may be included and thus weigh on the nature of the trip. On every stop after the first, a next stop purpose decision is made. One of the alternatives in this decision, along with stopping at a client (business) and stopping at a non-client, such as a gas station or restaurant (other) is to make the decision to return to the establishment. A diagram of this is shown in figure 2. The iterative loop continues until the next stop purpose model determines that the vehicle is returning to the establishment, at which time the final trip is known and the tour is finished. The microsimulator then continues to the next generated tour. The primary source of data for this model is an extensive set of surveys of the shipments arising and associated vehicle movements made by business establishments – analogous to a household trip diary – collecting information on tours made in the Calgary region by just over 3,000 establishments on a typical weekday in 2001 (10). Sampled establishments provided information on the movements of their entire fleet over a 24 hour period, including origin, destination, purpose, fleet and commodity information. Information on over 64,000 trips was available for the estimation and calibration of the tour based model. MODEL DEVELOPMENT Model Fundamentals The Calgary CVM operates based upon three vehicle classes; light, medium and heavy. Light vehicles are passenger vehicles (cars, vans, pickup trucks), and operate over the entire road network. Medium vehicles are single-unit vehicles with six tires, and heavy vehicles are multi-unit vehicles with more than six tires. Both medium and heavy vehicles are subject to truck route restrictions in Calgary. Truck routes are explicitly modelled in the Calgary CVM; medium and heavy trucks are forced, through network penalties used in assignment, to travel only on truck routes unless making a stop that is inaccessible from truck routes. When travelling off of truck routes, they are to take the shortest possible path. Truck route travel paths are used in the calculation of travel times, generalised costs and accessibilities, and these were used throughout estimation. As a result, medium and heavy vehicles ‘see’ a different network from light vehicles, and truck route policies can be tested. A concept of purpose is also included in this model, much as ‘work’ and ‘school’ are included in a household personal travel model. These purposes represent different types of activities, with different influences and choice structures. The four purposes are goods, which includes handling goods; service, which is the performance of service (or goods handling to perform a service, such as a plumber picking up supplies); other, which comprises all non-business purposes; and return, which represents the return trip to the business establishment, either at the end of the day or during the day for any reason. Companies are segregated into 5 industry categories: private services (which includes government and education), retail, industrial (which includes agricultural), wholesaling and transportation. Each of these industry categories has unique coefficients throughout much of the model, so the model produces very different behaviours and reactions to policy changes for these different categories. The model also has a slightly different structure for ‘transportation’, most notably in that the goods and service stop and tour purposes are combined into a single business purpose, because transportation firms provide the service of handling goods, which blurs the definitions. Tour Generation Model The aggregate number of tours generated by each industry category is determined first for each time period in each model zone. These aggregate numbers of tours are used to form lists of discrete tours considered in the rest 4 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal. of the model. Tour generation rate (tours per employee in an industry) is determined using a regression with zonal attributes such as land use and accessibility included as independent variables. This rate is multiplied by the number of employees in the relevant industry in the zone to produce a total number of tours generated. One interesting result from the parameter estimations undertaken in the development of these regressions was the strong tendency for accessibility to have a negative constant. This differs from the expected positive value for models of personal travel. We believe this is due to differences in the motivations for commercial establishments and households. If congestion increases, a company’s vehicles will take longer to make their deliveries. At least in the short term (which is what the model covers) the company is still required to fulfil the needs of its clients, so it will have to add vehicles to the fleet and make more tours to be able to make the same number of deliveries. Tour Purpose and Vehicle Choice Model Each tour is assigned both a primary purpose and a vehicle type simultaneously. This is done using a Monte Carlo process where the selection probabilities are determined using single-level logit models based on industry type with utility functions that include land use, establishment location and accessibility attributes. The alternatives for the primary purpose are to (a) handle goods, (b) perform service visits or (c) undertake solely other non-business purposes. The alternatives for vehicle type are (a) light, (b) medium or (c) heavy. The need for a tour purpose model arose from earlier work, which revealed that virtually all tours make goods stops or service stops, but not a combination of both. Attempts to develop a nested model of these two choices resulted in nesting coefficients that were either unacceptable (greater than 1) or not significantly different from 1, indicating that the two decisions can be considered appropriately with a single-level structure. The generalized utility function for tour purpose and vehicle choice is as follows: Utour purpose, vehicle = ASCtour purpose, vehicle + purpose land use + vehicle land use + purpose company + vehicle company + population accessibility vehicle, land use × population accessibility for generating zone + employment accessibility vehicle, land use × employment accessibility for generating zone where ASCtour purpose, vehicle is the alternative specific constant for a given combination of tour purpose and vehicle choice (the constant for a goods tour by medium vehicle, for instance, is –5.393). purpose land use and land usevehicle are constants for the vehicle choice and tour purpose, based on establishment zone land use. For instance, in an industrial land use zone, the constant for a goods tour by medium vehicle is adjusted by –0.1517 and +2.391, respectively. purpose company and companyvehicle are similar to land usepurpose and land usevehicle, but based on the industry of the company generating the tour. For instance, for a wholesaling company, the constant for a goods tour by medium vehicle is now adjusted by +3.946 and +0.4309, respectively. vehicle, land use population accessibility and employment accessibilityvehicle, land use are based on the land use of the establishment zone. For instance, for a medium vehicle in an industrial zone, the population accessibility is multiplied by 9.866×10-6. Coefficients for the above generalized function are shown in tables 1 through 3. Table 1 has the values for ASCtour purpose, vehicle, table 2 has the values for land usepurpose, land usevehicle, companypurpose and companyvehicle and table 3 has the values for population accessibilityvehicle, land use and employment accessibilityvehicle, land use . The tour purpose and vehicle choice model for the transportation industry is a reduction of the above generalized utility function. Transportation establishments were modelled with a single purpose tour, so only vehicle choice needs to be covered and the constants relating to tour purpose are not necessary. The resulting coefficients are shown separately for transportation establishments in table 4. 5 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal. The first five terms of the utility function shown are simple constants. Figure 3 shows, using only the constants and neglecting accessibilities, the proportions of vehicle and tour purpose produced for a wholesaling company based in each of the land use categories. A wide variation can be seen in both vehicle type and tour purpose; for instance, there is a big difference in the use of light and medium vehicles for establishments based in residential versus commercial/retail land uses. Some ceritas paribus general trends indicated by the estimated values are notable. Overall, light vehicles are used more than medium and heavy vehicles. Retail companies are even more likely to use light vehicles, whereas industrial, wholesaling and transportation companies are more likely to use medium and heavy vehicles. As expected, companies based in residential areas are most likely to use light vehicles, with commercial/retail land use emphasizing medium vehicles more than the other land uses. Similarly, service is favored compared to goods, with other tours a relative rarity. Unsurprisingly, private service companies are more likely to make service tours than other companies, with wholesaling then retailing having the highest propensity to make goods tours. While industrial companies are more likely and the wholesaling less likely to make other tours, nobody makes a lot of them. Land use also plays a role in this process, with commercial/retail the most likely to produce service tours and low-density the most likely to produce goods tours. The accessibilities provide some feedback of network conditions into this model, although they have minor effect. It can be argued that this is true in reality – even very large changes in network congestion would be unlikely to bring about changes in vehicle choice for a flower delivery service or a gravel hauler. Tour Start Time Model Tours are first generated within five time periods, divided from the daily total using a single-level logit model. The Calgary CVM deals with exact time; each stop occurs at a specific time, so the model output can be aggregated in different time periods as necessary. The model uses a Monte Carlo process is used to assign an exact start time, with sampling distributions based on observed start times differentiated by industry and time period. Next Stop Purpose Model The purpose is selected for each subsequent stop, from the alternatives goods, service, other or return to establishment. If the primary purpose of the tour is to handle goods, then the service alternative is not available for stops on the tour, and vice-versa. If the primary purpose is to undertake other purposes, then neither the goods nor the service alternatives are available. The term business stop refers to either the goods purpose or the service purpose as appropriate. If the purpose selected for the subsequent stop is return, then the next stop location is already known, the final trip is made and the microsimulator moves on to the next tour. This return purpose is not available on the first trip. Figure 2 illustrates a typical next stop purpose decision. Consideration of next stop purpose, along with the other ‘iterative’ elements of the model (including next stop location and stop duration) is divided into 13 ‘segments’, where a segment combines an establishment industry, a vehicle type and a tour/stop purpose. (The next stop purpose model uses the tour purpose; the other two use stop purpose.) These 13 segments were chosen based on the preliminary estimation results and in order to ensure a suitable amount of estimation data for each segment. The generalized utility functions are as follows: Ubusiness = business previous × ln(number of previous business stops) + ASCbusiness Uother = other previous × ln(number of previous other stops) + other elapsed time × total elapsed time + ASCother Ureturn = total previous × ln(number of previous stops) + elapsed time ×total elapsed time + travel time × total travel time + return gen cost × generalized cost for return to establishment + ASCreturn 6 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal. Coefficients for the above are shown in table 5. The alternative specific constants indicate that other stops are universally not very common, except on other tours, and that tours tend to perform primarily business stops. The previous stop coefficients show that as a tour continues the propensity to make additional stops increase. However, the elapsed time and travel time coefficients will increase as additional stops are generated, pulling vehicles back to their establishments. In order to make a large number of stops, each stop must be made quickly – which is confirmed in the data, where security companies were found to visit 50 or more locations in a night, stopping for a minute or two at each. The total elapsed time includes the total travel time, so the positive values for the coefficient for total travel time indicates that the travel time component of the elapsed time has a further positive influence on the tendency to return to establishment. This suggests that travel time and the associated locations of clients being served in tours from establishments has more of an influence on the structure of tours than does the time spent at the clients. The influence of total elapsed time includes the influence of workshift patterns, with lunch break and quitting time occurring at fixed times some 4 and 9 hours of total elapsed time after the start of the workday. With hindsight, it is seen that additional variables indicating the differences in time to lunch break (perhaps noon) and quitting time (perhaps 4:00pm) would have been useful in helping provide a more complete representation of the influence of such factors along with the general tendency for tours to end more directly. Next Stop Location Model Once the purpose of the next stop is selected, assuming it is not return to establishment, then the location of the next stop is selected. Again, a Monte Carlo process is used with the selection probabilities – across the full set of 1,447 model zones – determined using a logit model with a utility function. The generalized utility function for travel from zone i to zone j is as follows: Ulocation(j) = land use + average income × zone j average household income + gen cost of travel× generalized cost of travel from current zone to zone j + return gen cost × generalized cost of travel to return to establishment from zone j + population accesib × population accessibility + employment accesib × employment accessibility + enclosed angle × enclosed e-i-j angle + size term × ln (zone j population + relative employment size term × zone i employment) where is chosen based on the land use of zone j; based on the five land use types (low density, residential, land use commercial/retail, industrial and employment node), the value of land use is set to 0 for residential. enclosed e-i-j angle is the angle enclosed by the lines connecting the establishment and zone i, and zone i and zone j. This angle is measured in degrees, so a value of 180° represents travel directly away from the establishment, and a value of 0° represents travel in the direction of the establishment. The last part of this function is slightly different for transportation companies, where size term is not multiplied by zone population or by employment, but by the output of a separate visit regression model. This visit model simulates demand for transportation firms, and is based on the employment of a zone, the accessibility of the zone, the population of the zone, and the proportion of employment in various industries. It is included in this case in recognition of the significant role of transportation depots and the resulting appropriateness of increasing the accuracy of representation regarding their attraction of commercial vehicle trips. The coefficients estimated for the location choice model are shown in table 6, with the 0 value for residential land use omitted because of space restrictions. Tours are generally ceritas paribus most attracted to industrial areas, and least attracted to employment nodes. Industrial areas tend to house warehouses, production sites and other types of activity that need more goods provided. 7 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal. The most significant value in the estimations is the generalised cost of travel; t-ratios were typically in the 30-50 range. That commercial vehicles are cost-averse is no surprise. The role of return generalised cost was smaller, however it was significant, indicating that commercial vehicles do consider how far they are straying from base. The accessibility coefficients were generally less significant, and the negative signs are likely the result of correlations. A negative coefficient for enclosed angle indicates a tendency to circle back towards the establishment; however, there is a strong correlation with the return generalised cost, and it is worth noting that some of the positive values for this coefficient are in segments with high value of return generalised cost. The size terms have a strong impact as well, with different segments being attracted to employment and population in different ratios. Stop Duration Model The stop duration model is similar to the tour start time model. The duration of each stop being made by the vehicle is established and used to advance the clock keeping track of the start and end times. A Monte Carlo process is used to determine the duration of the stop, with sampling distributions based on observed durations. There are different duration distributions for each of the 13 ‘segments’; for instance, an other stop will tend to be shorter than a private service–service–light vehicle stop. The microsimulator then returns to the next stop purpose module and iterates. Model Calibration Following the estimation process, a series of calibrations were performed. The elements of the model are interdependent; for example, if the tour generation is adjusted, establishment locations are changed, which affects the decision to return to establishment and therefore tour lengths. This interdependence necessitates an iterative calibration process where several different elements are adjusted simultaneously. The model was calibrated to match five sets of targets: • Tour generation by industry and geographic area (within the generation model); • Tour length (adjusting ASCreturn in the next stop purpose model) ; • Vehicle type and tour purpose proportions (adjusting a set of ASCtour purpose, vehicle for each company type); • Total trip destinations for 13 super-zones, including intra-zonal proportions (adding a set of k-factors to the next stop location model); and • Number of trips within the AM, PM and combined offpeak periods (adjusting the tour start time model). Fits were within 5% of observed values in all cases, and many are within a fraction of a percent; most of the coefficient changes are fairly small. Typical values for the added ASCtour purpose, vehicle are ±0.5, for instance. Model Capabilities The elements in the Calgary CVM permit a wide variety of responses to network changes. This enables not only the realistic simulation of commercial vehicle movements, but the testing of a number of possible scenarios. This model will respond to changes in areas such as: • Road network capacities and connectivity • Personal travel (congestion as a result of the personal travel model) • Truck routes • Tolls • Fuel taxes • Population and employment sizes and distributions (and associated growth and development) • Employment densities • Employment types (economic shifts between industries). The above changes will be responded to in most levels of the model. If travel becomes more onerous – if the network becomes more congested, or if a key truck route is removed – then vehicles will not just travel shorter distances. They will also make shorter tours, and more tours will be generated to fulfil demand. Tour generation, 8 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal. start time period, tour purpose, vehicle type, stop purpose and stop location are all influenced by network conditions, and all of them will respond to changes such as the ones listed above. CONCLUSIONS The nature of urban commercial vehicle movement is not widely understood; the principles governing larger regional models are often incorrectly assumed to apply to urban areas. This discounts many important elements, including the importance of trip chaining, the significance of service delivery, and the role of light commercial vehicles. The development and use of models based on the freight-only, large-truck view of commercial movements in urban areas leads to the neglect of the half of commercial trips made in light vehicles and the half of trips made for service delivery. Urban commercial movement is different from both household movement and regional commercial movement, and thus requires different modelling. The urban commercial movements model described here is a successful demonstration of several principles. The generation model responds to network conditions in a realistic manner. The use of constants in the tour purpose and vehicle choice model still permits a broad spectrum of behaviours, and a reaction to changes in employment and population. A tour-growing model permits all stops on a tour to have importance, and permits tours of highly variable lengths. The interactions between each model component and the information from the network mean that changes to the modelled system affect the entire behaviour of commercial vehicles. Some areas of the model have potential for improvement; particularly the stop duration model. Additionally, the formulation of a more advanced fleet allocator model – perhaps based on a similar tour-based microsimulation or some form of use of the Clarke-Wright dispatch algorithm (11) rather than a gravity model – will complement the current tour-based microsimulation model. A more complete spatially disaggregated input-output land use and transport model would also provide a richer modelling environment to inform the decisions made by the CVM. Overall, the tour-based microsimulation approach has proven to be very successful: it delivers stable and robust answers, calculates quickly, gives great flexibility and permits very subtle influences and interactions to be modelled. The parameter estimations have provided great insight into the decision-making of urban commercial vehicles, and this model permits a good representation of this usually neglected component of urban traffic. REFERENCES 1. Ogden KW, 1992, Urban Goods Movement. Ashgate, UK. 2. Harris RI and Liu A, 1998, Input-output modelling of the urban and regional economy: the importance of external trade. Regional Studies 32(9):851-862. 3. Ortúzar JdeD and Willumsen LG (1994) Modelling Transport; Second Edition. Wiley, New York NY, USA. 4. Cambridge Systematics, Comsis Corporation and University of Wisconsin at Milwaukee (1996) Quick Response Freight Manual. Report DTFH61-93-C-00075. Prepared for the United States Federal Highway Administration, Washington DC, USA. 5. Boerkamps J, van Binsbergen A and Bovy PHL (2000) Modeling behavioral aspects of urban freight movements in supply chains. Transportation Research Record 1725:17-25 6. Hunt JD and Abraham JE (2003) Design and application of the PECAS land use modelling system. Proceedings of the 8th International Conference on Computers in Urban Planning and Urban Management, Sendai, Japan, May 2003, CD-Rom Format. 7. Jonnalagadda N, Freedman J, Davidson WA and Hunt JD (2001) Development of a microsimulation activity-based model for San Francisco: Destination and mode choice models. Transportation Research Record 1777:25-35. 8. Slavin HL (1998) Enhanced framework for modeling urban truck movements. Proceedings of the 6th National Conference on Transportation Planning for Small and Medium-Sized Communities. (http://ntl.bts.gov/lib/000/500/571/00780105.pdf) 9 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal. 9. Slavin HL (1979) The Transport of Goods and Urban Spatial Structure. PhD Dissertation. Cambridge University, UK. 10. Hunt JD and DG Morgan (2001) The Calgary Region commodity flow survey. Proceedings of the 2001 Annual Conference of the Canadian Institute of Transportation Engineers, Calgary AB, Canada, May 2001. 11. Clarke G and Wright JW (1964) Scheduling of vehicles from a central depot to a number of delivery points. Operations Research 12:568-581. 10 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal. LIST OF TABLES AND FIGURES TABLE 1 Tour Purpose and Vehicle Choice Alternative Specific Constants TABLE 2 Tour Purpose and Vehicle Choice Land Use and Company Type Constants TABLE 3 Tour Purpose and Vehicle Choice Accessibility Coefficients TABLE 4 Tour Purpose and Vehicle Choice Transportation Company Coefficients TABLE 5 Next Stop Purpose Coefficients TABLE 6 Next Stop Location Coefficients FIGURE 1 Process flow of Calgary CVM. FIGURE 2 Hypothetical tour growing illustration. FIGURE 3 Proportions of tour purposes and vehicle choices for a wholesaling company establishment located in different land use zones. TABLE 1 Tour Purpose and Vehicle Choice Alternative Specific Constants Light vehicle Medium vehicle Heavy vehicle Service tour 0 -3.944 -4.466 Goods tour -3.053 -5.393 -5.700 Other tour -3.705 -7.481 -9.620 All values significant at the 99.9% level (t-ratio > 3.29) TABLE 2 Tour Purpose and Vehicle Choice Land Use and Company Type Constants Light Medium Heavy Service Goods tour Other tour vehicle vehicle vehicle tour Low-density land use 0 2.047 2.461 0 .3460 1.603 Residential land use 0 0 0 0 0 0 Commercial / retail land use 0 5.023 3.145 0 -.9187 .6176 Industrial land use 0 2.391 2.144 0 -.1517 1.012 Employment node land use 0 2.861 1.652 0 .01522** .7474 Private service company 0 0 0 0 0 0 Retail company 0 -1.529 -2.386 0 3.105 -.04871** Industrial company 0 .5347 .9954 0 2.018 .4485 Wholesaling company 0 .4309 .3835 0 3.946 -.2193** * indicates significance at the 95% level (t-ratio > 1.96) but not at the 99.9% level (t-ratio < 3.29) ** indicates significance at less than the 95% level (t-ratio < 1.96) All other values significant at the 99.9% level (t-ratio > 3.29) 11 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal. TABLE 3 Accessibility Coefficients affecting Vehicle Choice in Tour Purpose and Vehicle Choice Model Low-density Residential land Commercial / Industrial land Employment land use use retail land use use node land use Light vehicle 4.207×10-5 0 4.860×10-5 2.262×10-5 3.512×10-5 population accessibility Light vehicle 0 0 1.027×10-4 0 0 employment accessibility Medium vehicle 0 0 0 9.866×10-6 0 population accessibility Medium vehicle 0 0 9.705×10-5 0 1.414×10-5 employment accessibility Heavy vehicle 1.356×10-4 0 4.891×10-5* 0 1.736×10-6** population accessibility Heavy vehicle 0 0 1.057×10-4 0 0 employment accessibility * indicates significance at the 95% level (t-ratio > 1.96) but not at the 99.9% level (t-ratio < 3.29) ** indicates significance at less than the 95% level (t-ratio < 1.96) All other values significant at the 99.9% level (t-ratio > 3.29) TABLE 4 Coefficients for Transportation Companies in Tour Purpose and Vehicle Choice Model Light vehicle Medium vehicle Heavy vehicle Alternative-specific constant 0 .7417 1.897 Low-density land use 0 .02558** -.1049** Residential land use 0 0 0 Commercial / retail land use 0 0 .6191* Industrial land use 0 0.5390 -.2808* Employment node land use 0 -.2192** -2.163 Employment accessibility 0.1268×10-4 0 0 * indicates significance at the 95% level (t-ratio > 1.96) but not at the 99.9% level (t-ratio < 3.29) ** indicates significance at less than the 95% level (t-ratio < 1.96) All other values significant at the 99.9% level (t-ratio > 3.29) 12 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal. TABLE 5 Next Stop Purpose Coefficients Company types / Tour ASC ASC Other ASC Business Other Total Elapsed Travel time Other Return Employ. purpose / Vehicle Business Return previous previous previous time (×10-3) elapsed gen. cost accessib. type(s) time (×10-2) All / Other / All n/a 0 4.083 n/a 0 -3.380 .7893 0 0 26.96 7.015×10-7** Private service / 2.936 0 2.639 .3514* .2715** -1.045 .2539 5.969 .1046* 3.981 0 Service / Light Private service / 2.352 0 2.162 .4774 1.053 -.7774 .3402 2.587 .1048 6.057 0 Service / Medium and heavy Private service / 2.284 0 1.648 1.133 1.336 -.5174 .3909 6.431 .2716 1.106** 0 Goods / All Retail / Service / All 2.707 0 2.619 .6021 .9202 -.1112** .1837 -.8995** .1532 5.538 0 Retail / Goods / All 3.725 0 3.411 .1141** 1.557 -1.519 .2083 8.930 -.1128** -3.348 0 Industrial / Service / 2.525 0 2.978 1.075 1.121 -.9242 .3525 3.123 .2234 3.253* 0 Light Industrial / Service / 2.599 0 2.364 6.148 1.202 -1.133 .3025 9.960 .1187 10.75 0 Medium and heavy ×10-2** Industrial / Goods / 2.890 0 3.041 .3996 .9585 -1.127 .2748 4.555 .1103 3.335 0 All Wholesale / Service / 2.302 0 2.028 .9692 1.159 -.3461* .3419 2.754** .1509 9.744 0 All Wholesale / Goods / 3.448 0 1.823 .4821* 1.412 -.4929* .2715 4.501* -.1719 1.402** 0 Light Wholesale / Goods / 2.984 0 1.687 .3894 1.316 -.4665 .1746 10.28 6.591 2.118** 0 Medium and heavy ×10-3** Transportation / n/a / 2.901 0 2.541 1.395 2.174 6.366 .2944 1.819 .2447 7.048 0 All ×10-2** * indicates significance at the 95% level (t-ratio > 1.96) but not at the 99.9% level (t-ratio < 3.29) ** indicates significance at less than the 95% level (t-ratio < 1.96) All other values significant at the 99.9% level (t-ratio > 3.29) 13 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal. TABLE 6 Next Stop Location Coefficients Company types / Low Comm. / Industrial Employ. Average Gen. cost Return . Population Employ. Enclosed Size term Relative Tour purpose / density retail land land use node land income of travel gen. cost accessib. accessib. angle employ. Vehicle type(s) land use use use (×10-6) (×10-6) (×10-6) (×10-3) size term All / Other / All -.7902 .02702** -.1595* -.6126 -11.49 .3039 .1310 -7.651 -9.696 -2.346 .2800 6.779 Private service / -.08976* -.2755 .2152 -.4623 1.676* .3283 0 -10.83 -2.653 -1.884 .3094 .08658** Service / Light Private service / .7250 -.1057** .4655 -.7546 9.476 .08481 .1229 -44.65 9.296 3.684 .2219 0 Service / Medium and heavy Private service / -.3327* .5674 .4926 .2062 0 .5688 0 5.717* -16.54 -6.348 .1588 0 Goods / All Retail / Service / -.9676 -.2310 .1547* -.5132 0 .3601 .03662 -17.35 0 -1.241* .2841 .6334 All Retail / Goods / All -.1707 -0.0256** .8014 -.1840 0 .3734 .09158 -13.32 -1.682* 1.914 .2067 1.633 Industrial / Service -1.144 -.2361 .05026** -.4182 0 .2869 0 -24.98 5.477 -3.067 .2371 1.231 / Light Industrial / Service 0 -.3231* .2789* -.8438 0 .1627 .1279 -13.25 -16.96 2.934 .1205 1.012** / Medium and heavy Industrial / Goods / 0 -.1497 .5575 -.2042 0 .2581 .09615 -10.68 -5.139 -2.146 .2722 0 All Wholesale / -.9340 -.2130* .1440** -.4410 2.367** .3849 .04300* -11.81 -27.71* 1.761 .2426 2.138 Service / All Wholesale / Goods -.6668 0 .9271 -.2688 0 .4495 .1075 -32.84** -67.74 .8923 .2248 0 / Light Wholesale / Goods -.1226** 0 .1445* -.1183* 0 .3123 .03430 -31.84 5.950 -1.431* .3021 2.313 / Medium and heavy Transportation / n/a -.5279 .1004 .6275 .02668** -4.691 .3792 0 -11.72 -5.984 3.109 8.651×10-2 n/a / All * indicates significance at the 95% level (t-ratio > 1.96) but not at the 99.9% level (t-ratio < 3.29) ** indicates significance at less than the 95% level (t-ratio < 1.96) All other values significant at the 99.9% level (t-ratio > 3.29) 14 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal. Tour Generation Vehicle and Tour Purpose Tour Start Next Stop Purpose Next Stop Location Stop Duration FIGURE 1 Process flow of Calgary CVM. Client Non-client (e.g. Gas Station) Establishment Client Client Non-client (e.g. Restaurant) Client FIGURE 2 Hypothetical tour growing illustration. 15 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal. Stefan, McMillan and Hunt 16 100% 90% 80% 70% Heavy other Proportion of generated trips Heavy goods 60% Heavy service Medium other 50% Medium goods Medium service Light other 40% Light goods Light service 30% 20% 10% 0% Low-density land use Residential land use Commercial / retail land Industrial land use Employment node land use use FIGURE 3 Proportions of tour purposes and vehicle choices for a wholesaling company establishment located in different land use zones. 16 TRB 2005 Annual Meeting CD-ROM Paper revised from original submittal.