The CCMPO travel demand model is expected to be used primarily for the following purposes:
The CCMPO uses the integrated land-use and transportation model regularly for long-range planning and corridor studies. The regional transportation plan is updated every three to five years and the model is a valuable tool in that process.
The model has been used for a broad range of applications, including allocating future land-use growth, understanding the traffic impacts of large developments, as well as analyzing the merits of enhancements to the existing transportation system and new roadway and transit facilities.
After introducing the history of the travel demand model, and its current uses, CCMPO and the agency's primary modeling consultant Resource Systems Group, Inc. (RSG) presented information about the inputs to the model and each of the individual model components currently in use. The following sections summarize the information provided by CCMPO staff, as well as comments from peer review participants. CCMPO provided the peer review panel with model documentation. The "CCMPO Regional Travel Model Documentation Version 3.2" describes all of the model components and input data and presents a summary of model validation results.
The CCMPO travel demand model covers all of Chittenden County in the northwestern corner of Vermont. For modeling purposes, major roadways within the modeling region were selected to reliably represent the entire road network. Within the model boundary, the network consists of all roads that have a federal functional classification above local streets as well as some roads that are not classified but that are important for network connectivity. There are over 1,800 road segments represented as links, of which approximately 200 are one-way only.
The network is based on a GIS file of centerlines with various network attributes that describe the roadway characteristics. The network has posted speeds and hourly capacity. The hourly capacity is based on the hourly ultimate capacity, that is, the point at which the Level of Service (LOS) changes from an "E" to an "F" as defined by the Highway Capacity Manual.
Capacity varies by functional class, presence of turn lanes, and the number of lanes. The model includes delay from both links and intersections using a logit-based volume delay function which calculates a link and node contribution to total delay using assumed link and node capacities and vehicle flows.
Transit routes are coded into the model based on information obtained from the Chittenden County Transportation Authority (CCTA), including routes, fares, and headway information. Because stops along a route are likely to change frequently, the current physical route stops are not modeled precisely. Instead, a stop was placed on nodes along the route's path as appropriate to represent the service provided by each route. All links from the roads layer are included as non-transit links in the transit network, allowing travel by foot, bike, or car (to rail only) from anywhere in the network to any transit route. The transit network includes bus routes and the option to include rail routes although no rail service currently exists in the region.
The CCMPO travel demand model, as is typical, uses traffic analysis zones (TAZ) as the base geographic unit. CCMPO forecasts the number of employees (retail and non-retail), and the number of households to determine the number of trips generated in each zone. The CCMPO model has 335 TAZs (not including external station TAZs), which approximates a modeling guideline of 0.6 TAZs per square mile. TAZs are derived from census tracts and block groups, some of which are subdivided as necessary. In creating TAZs, CCMPO's goal was to represent how traffic enters and exits a particular TAZ. Therefore, major roads or other features that create barriers between adjacent land uses are normally used as TAZ boundaries.
CCMPO prepares their base year household location information using housing data collected by the Chittenden County Regional Planning Commission (CCRPC) based on the 2005 municipal 'Grand List'. The data was then compared to the available parcel data, the 2000 Census data, building permits, and a random windshield survey. Based on these comparisons, the 2005 Grand List was determined to be the most accurate data source.
The housing data includes the total number of households per TAZ, classified by household size (0,1,2,3,4+) and auto ownership (0,1,2,3+). The household size and auto ownership distribution assumptions come from the 2000 Census, at the Census tract level.
CCMPO and CCRPC collected employment land use data from two distinct sources: infoUSA (a commercial data provider) and the Vermont Deptartment of Employment and Training (DET). Since the VT DET employment has a privacy agreement and use restrictions, the CCMPO chose to use the infoUSA data and supplemented gaps in the infoUSA data using the VT DET data. The infoUSA data contains information such as the name of the employer, the address of the employer, the general number of employees, and the employer's Standard Industrial Classification (SIC) code.
Based on the addresses of the employers, CCMPO was able to geocode more than 90 percent of the total employers. For those addresses that were not easily geocoded, CCMPO focused more resources to those employers with more than 5 employees. Based on this effort, CCMPO was able to geocode more than 98 percent of the total employees included in the dataset.
Once the employment data was geocoded, each employer could be assigned to one of the 335 internal transportation analysis zones (TAZs).
Employment was then disaggregated into nine employment categories created to reflect particular trip attraction similarities: Accomodations, College, Commercial, Industrial, Institutional, K-12 School, Retail, Special Commercial, and Special Retail.
CCMPO develops control totals for housing, and employment in the future year (2030) for the entire CCMPO region and then allocates these to the TAZs based on the availability of land for development and the attractiveness of the TAZ based on a number of factors. CCMPO uses their own LUAM, a model developed by RSG that has been calibrated to be reflective of local development patterns.
The purpose of the land use allocation model (LUAM) is to create land use scenarios that are realistic, based on land policies in effect, internally consistent with the transportation system, and can be easily updated. These future transportation/land use scenarios must also be realistically influenced by transportation measures including transit improvements and land use policy decisions. The land use allocation model is used to allocate user defined land use control totals of county-wide housing and employment to the transportation analysis zone (TAZ) structure.
The allocation process generally takes three effects in to account: 1) The availability of land in each TAZ specified in the allowable land use file, 2) the accessibility of each TAZ (composite impedances) calculated by the mode choice model of the transportation model, and 3) the existing land use already in place specified in the existing land use file.
The LUAM was initially calibrated for the 1998 model. The calibration was revisited as part of the development of the new 2005 base year daily model. RSG used historical parcel data, 1990 Census data, and the observed 2005 land-use to assemble a 1990 land-use dataset for Chittenden County. The 1990 dataset was then used as an input to the LUAM to forecast land-use in 2005. The forecasted land-use was then compared to the observed land-use in 2005 at an aggregate level. The parameters of the LUAM were then modified so the forecasted data matched the observed 2005 data more closely.
Panel Discussion
The panel discussed the merits of adding a vehicle ownership model to the regional travel demand model. In the current model, the auto ownership distribution is extracted from Census data and zonal shares by auto ownership category are held constant in future analysis years.
An auto ownership sub-model would estimate probability distributions for four different choices: own zero, own one, own two, or own three or more vehicles. The panel suggested one principal advantage to incorporating an auto ownership model is the added sensitivity to urban form variables that could be realized. In addition the panel suggested it would be relatively easy and straight forward to implement a borrowed logit model structure and parameters from other regions that have successfully implemented vehicle ownership models.
The Trip Generation model estimates the number of trips that each TAZ produces or attracts, and CCMPO implements this process within the TransCAD model structure using a series of GISDK scripts.
The CCMPO model uses a cross-classification approach to estimating trip productions. This approach is widely used in other regional network models. For each TAZ, households are cross-classified according to size (1 person, 2 persons, 3 persons, or 4+ persons) and auto ownership (0 auto, 1 auto, 2 autos, or 3+ autos). The cross-classification approach requires that separate trip production estimates be developed for each of the resulting 16 household types. The production rates were estimated using the 1998 Chittenden County household survey and the 2001 NHTS survey. CCMPO purchased the Add-On data for the 2009 NHTS which was also used to check and update the trip production rates.
Trip productions are estimated for three internal person trip purposes, Home-based work (HBW), home-based other (HBO), and nonhome-based (NHB).
Trip attraction models for HBW, HBO and NHB trip purposes were developed originally from the CCMPO household diary survey. Commercial trip purposes are also included so as to explicitly account for the commercial trip demand which makes up a significant share of total daily travel. Light duty vehicles (4-tire), Medium (6-tire) and Heavy (> 6-tire) commercial trip rates came from FHWA's Quick Response Freight Manual. The estimated model coefficients were then calibrated so that trip attractions by purpose would be consistent with trip productions and proportionate to ITE trip generation rates by land use type.
Panel Discussion
The panel suggested that CCMPO consider adding more market segmentation to the Trip Generation module by adding more trip purposes and more employment categories.
The panel recommended adding a home-based shopping (HBSh) trip purpose which is currently represented by the home-based other purpose. The region does have distinct retail centers and therefore representing shopping trips explicitly could improve the model.
The panel also suggested adding home-based school (HBSch) and a home-based university (HBUniv) trip purposes. Given the region is home to a number of colleges and the University of Vermont most notably among them, a university purpose has been considered in the past. The panel described HBUniv models that rely on student and employee zip-code information collected from the institution that can be used instead of applying a gravity model.
The panel stressed the importance of disaggregating these purposes from a basic home-based other trip purpose due to unique time of day patterns and mode choices associated with these trip purposes.
Finally, beyond including more trip purposes, the panel recommended utilizing a different cross-classification scheme using either workers and/or income in place of auto ownership in the trip production models. Income and presence of workers might be a better predictor of household trip-making than whether or not the household owns an automobile. In the attraction models, the panel suggested using two-digit NAICS employment codes which would result in more detailed employment types beyond the nine employment types currently utilized in the model.
The CCMPO travel demand model contains 17 external stations. The most significant external stations are located at the Interstate 89 points of entry/exit which is the only interstate in the model region.
In the base year, the total internal-external (IX), external-internal (XI) and external-external (XX) trips are set to match base year traffic count data at the external stations. The percent of XX trips is derived in large part from an external license plate survey, while the remaining external trips are split among IX and XI trips based on the AM and PM directional imbalance observed in traffic counts as a proxy for the home-end (production) locations of the trips. The model assumes for most external TAZs that the IX and XI trips will be 35 percent internal-to-external (IX) and 65 percent external-to-internal (XI), meaning approximately 2/3 of the IXXI demand will be generated externally (i.e. by people who live outside the region). These assumptions are allowed to vary by external TAZ location and do on Interstate 89 south of the region, where the split is 55 percent XI and 45 percent IX. . These assumptions have been recently validated using NHTS data.
Future year external trips are assumed to grow annually at a user-specified rate, which can vary by external TAZ. In the model, the external trips are initially assumed to have a growth rate of 1 percent per year but this rate is adjustable.
Panel Discussion
The panel suggested that applying a 1 percent per year growth rate to establish future year external trips was somewhat arbitrary. The panel recommended developing a regression using historical count data along with a procedure that takes into account both historical counts and growth in population and employment for zones proximate to the external stations.
The trip distribution model matches the productions and attractions of each TAZ with productions or attractions from other TAZs using a gravity model. The intra-zonal travel times used in the gravity model are an average of the travel time to the three nearest TAZs. The gravity model uses an exponential gamma function to develop the friction factors. All trip purposes are doubly constrained with the exception of external trips which are singly-constrained to the external end. Different gravity model parameters are used for each trip purpose.
Non-motorized and transit trips comprise a small portion of the total trips in the CCMPO region. In 2009, the local transit provider, Chittenden County Transit Authority (CCTA), was averaging about 7,800 boardings per day. Roughly 90% of the trips in the CCMPO utilize the auto mode. However, the CCMPO Regional Travel Demand Model does include a mode choice component to split trips into non-motorized, auto, bus, and rail modes.
The CCMPO travel model includes a new pre-distribution non-motorized binomial logit model where non-motorized shares are related to residential density, employment density, and intersection density, which serve as proxies for urban form variables which tend to increase walking and bicycling activity. A subsequent post-distribution nested multinomial logit-form mode choice model is used to estimate the split among auto trips, bus trips, and rail trips, and there are separate mode choice models for work and non-work trips. There is currently no rail in the CCMPO region, so the rail mode has not been calibrated, though the model functionality is in place and is set to use the same choice parameters as bus transit. This is the current Federal Transit Administration (FTA) recommended practice for analyzing rail transit alternatives.
Panel Discussion
The panel had some concerns about the overall structure of the mode choice models. The panel questioned whether identifying the split of motorized and non-motorized trips before trip distribution with a binomial logit model and subsequently splitting the motorized trips into auto and bus (or rail) with a nested multinomial logit model after trip distribution might lead to some inconsistencies and obfuscate the relationship among the parameters utilized in each model.
CCMPO explained that previous versions of the mode choice model included the non-motorized (walk/bike) modes as a separate nest in the multinomial logit model applied after distribution. The new model was intended to be more sensitive to urban form variables and avoid cases where the walk/bike mode is only a competitive option when weighed against very short drive trips.
The panel also discussed whether the post-distribution mode choice module should be included within the Distribution-Assignment feedback loop. In the current CCMPO model, the Mode Choice module was deliberately not included in the feedback loop since transit usage in the region is low and this approach significantly reduces model complexity and overall runtime.
Finally, the panel discussed the merits of adding a school bus mode in the context of also adding a home-based school trip purpose in the trip generation module. The panel explained that bus routes need not however, be explicitly coded into the transit route layer.
The purpose of the assignment model is to locate a specific route along links and through intersections for every vehicle trip. The vehicle trips calculated in the mode split model, which are in the form of an origin/destination matrix, are "assigned" to the network based on a user equilibrium model. The trip table is then input to a user equilibrium model, which uses an iterative process to achieve a convergent solution in which no travelers can improve their travel times by switching to another route.
The assignment model includes travel delay from five sources:
In travel demand modeling, delay is typically considered a function of the ratio of volume to capacity (v/c). As v/c ratios near 1.0, the delays become more severe. The delays attributed to v/c ratios are based on the volume-delay function parameters. The CCMPO travel model vehicle assignment algorithm uses a logit based volume delay function available within the TransCAD software developed by the Israel Institute of Transportation Research and Planning (IITPR). The function has the characteristics of including both link delay as well as delay caused at intersections. The total delay on a link is calculated as the sum of the link delay and an estimated intersection delay.
Panel Discussion
The panel recommended that a relative gap closure of 0.0001 and maximum iterations of between 50 and 100 iterations be utilized in the iterative assignment process. The current vehicle assignment module uses a fixed number of 30 assignment iterations.
The CCMPO travel demand model includes a feedback loop from assignment to trip distribution and uses the method of successive averages (MSA) to average results from each iteration with the average of previous iterations to reach convergence. As noted earlier the mode choice models are not included in the distribution feedback loop.
Panel Discussion
The reasonableness of the convergence quit criteria being applied in the CCMPO model was reviewed. Both the change in link volumes and the change in trip zone-to-zone interchanges from one iteration to the other are examined in determining convergence. The panel thought the approach and criteria being applied in the CCMPO model were sound.
