This section summarizes the morning session presentation and discussion of the InSITE model structure, validation status, and remaining validation and sensitivity testing tasks. Specific comments from the panelists are summarized in Section 4.0.
The InSITE model covers both Baltimore and Washington planning areas, with the exception of Queen Anne County, which was added to the MPO after the model development had begun. The InSITE model consists of an activity-based model that simulates the resident intra-regional travel and aggregate models to represent trucks, external, and airport-related travel. The overall model structure is shown in Figure 3–1.
The population synthesizer and associated procedures produce a synthetic population with eight person types (see Table 3–1) as well as a work and non-work value of time and effective time coefficient per household. The time coefficient is used in the activity-based model components and the value of time is used in auto assignment to group trips into five value of time segments.
The ABM components are sequenced by long-term choices, daily activity pattern choices, mandatory tour choices, fully joint and individual non-mandatory tour choices, intermediate stop choices, work-based tour choices and finally trip-level choices. The model components and sequence are shown in Figure 3–2.
Outputs from the upstream components inform and constrain the downstream component choices. Logsums from tour mode choice are used in the tour destination and tour time of day choice models. To calculate tour mode logsums at the upstream level, intermediate results that have not yet been chosen, such as the number and purpose of intermediate stops, are temporarily simulated using distributions from the estimation dataset.
The following sections describe each of the activity-based model components in further detail.
Long-term choice components
The long-term choice components include the household-level auto ownership, transit-pass, and E-ZPass ownership components and the person-level school location and usual work location models.
School location is simulated for each child in the household. A child's choice of school location is sensitive to the location choice of younger children in the household. The sensitivity to a younger child's school location decreases as the difference in age between the children increases so that children close in age are more likely to attend the same school than children farther apart. The simulated school location is the default destination if the child makes any school tours.
The type and location of usual workplaces are simulated for each part-time and full-time worker in the region. The four workplace types that are simulated are: usual workplace in the region (not at home), usual workplace at home, usual workplace outside the model region, and no usual workplace. Unlike school location, the usual workplace is not necessarily the destination if the worker makes a work tour, but it is very likely to be. Work tour destination choice is simulated in the tour-level choice components.
Household class membership
Household class membership is a household-level model with six household class alternatives. The simulated household class influences the activity planning by household members for the day and represents the coordination between household members to plan joint activities.
Tour generation
Tour generation models are simulated at the person-level and create a roster of home-based tours for each person throughout the day. The tour types simulated are mandatory (work, school, university), non-mandatory (fully joint and individual), and the tours to escort children to/from school.
The school escort component simulates either the decision to escort a child to and/or from school as part of a mandatory (work or university) tour, individually as a stand-alone tour, or not at all. Children attending a school at the same location may be escorted together depending on a separate child-bundling process segmented by child age and half tour.
Individual non-mandatory and fully joint tour generation components simulate both the number of tours and the primary activity purpose (see Table 3–2 for a list of primary activity purposes). Fully joint tours are generated at the household level and a fully joint participation model operates at the person level to simulate the composition of each fully joint group. Only members of the same household can travel in a fully joint tour together and the entire tour is identical for each person, i.e. partially joint tours are not simulated. Joint travel and school escorting across households is not explicitly modeled, but it is not excluded. For example, an individual non-mandatory tour may use a shared-ride mode, implying that multiple persons are traveling together.
Tour level choice components
The tour level choice components simulate first the destination (for all except school tours which use the child school location simulated in the long-term components), time of day, the presence and purpose of intermediate stops by half-tour, and finally the tour mode.
For each tour purpose, the destination choice alternatives are the entire set of transportation analysis zones (TAZs) in the model network. Once a TAZ destination has been chosen, a parcel within the TAZ is selected based on the parcel population and employment characteristics.
Tour time of day alternatives are the tour arrival and departure half-hour period. The choice of tour arrival, departure, and duration are simultaneously simulated so the choice set is 49 * 48 / 2 = 1,176 to represent all possible half-hour time period pairs.
The number (between zero and three) and purpose of intermediate stops are simulated at the half-tour level (see Table 3–2 for list of intermediate stop purposes by tour type). Stops on the second (return) half-tour are sensitive to the number of stops generated on the first (outbound) half-tour.
The tour mode choice model is a 'shallow' implementation that leverages assignment and income segments to determine transit sub-mode and toll road usage. The tour modes are single occupant vehicle (SOV), high occupancy vehicle with two occupants (HOV2), high occupancy vehicle with three or more occupants (HOV3+), transit with auto access, transit with walk access, walk, bicycle, and school bus. Auto access to transit is not distinguished between park-n-ride and kiss-n-ride, but only transit stops that are coded in the network as park-n-ride have drive-access paths available so auto access is only simulated at those locations.
Work-based subtours
The tour generation and tour level choice model sequence is repeated for work-based subtours. These tours occur within a work tour and are simulated with the work location as the tour base location. The tour time of day is constrained to fall within the work tour time. The work-based subtour primary activity and intermediate stop purposes are shown in Table 3–2. The available tour modes are SOV, HOV2, HOV3+, transit with walk access, walk, and bicycle.
Trip level choice components
The trip level components include the intermediate stop destination choice, intermediate stop time of day, and trip mode choice.
The intermediate stop destination choice operates in a similar manner as the primary activity destination choice in that the choice set includes all TAZs and, once a TAZ is chosen, the parcel choice within the TAZ is simulated.
The intermediate stop time of day choice is constrained on one end by the primary activity tour begin or end time for stops on the first or second half tour, respectively.
The available trip mode choice alternatives are determined by the chosen tour mode choice. Trip mode choice also has constraints to ensure reasonable mode sequences, e.g. a drive-alone mode cannot occur after a transit or non-motorized mode on the first half-tour.
Trip are assigned through static processes. Transit trips are assigned in a production-attraction, peak/off-peak format. The highway assignment is segmented into eight time periods where each peak is represented with three periods. There are three truck classes and a separate class for the external and airport trips. The vehicle occupancy classes are further segmented into five sub-sets based on the value of time range and the median value of time from each range is used in the assignment.
The model validation process consists of the following steps:
At the time of the peer review, the single pass calibration step of model validation was completed through intermediate stop generation. Each model component has a spreadsheet of relevant comparisons to the household survey dataset with separate tabs containing different segments. The panel reviewed several of the validation spreadsheets from the completed components.
