The morning session of the peer review panel was spent by BCDCOG and their consultant presenting the model background and planning process. During these presentations, many topics came up which provoked discussion among peer review panel members and between panel members and BCDCOG. This section documents the key points that arose during these presentations.
The model was generally discussed in the order of execution, and this section is organized in a similar manner, beginning with the model inputs, progressing through trip generation, distribution, mode-choice, and assignment. Next, points from the discussion of the existing truck model and options to build a freight model, which was one of BCDCOG's primary areas of interest, are presented. This section concludes with the discussion notes from the model validation, user experiences, and data / tool integration questions.
BCDCOG has developed a socio-economic forecasting model in CommunityViz, but it is not fully integrated into the forecasting process. Instead, the modeler needs to manually translate the forecast developments into the model inputs using the spreadsheet method. There is no modeled connection between zonal accessibility and land use; instead BCDCOG uses their judgment to set growth in certain areas. An integrated land use-transportation model would consider the effects of accessibility on land use and location decisions. Several panelists commented that the connection of accessibility to land use is important because the traffic conditions have a strong impact on choices of land use and location decisions. This connection is not uncommon in most large urban area regional transportation models and is present in models of medium and small sizes as well. The panel suggested that BCDCOG rethink their model orientation so that it can demonstrate changes in the economic state of the region based on the transportation network performance, rather than only testing changes in the economic state on the transportation network.
The panel recommended that BCDCOG incorporate accessibility into their land use forecasts and integrate CommunityViz into the forecasting process.
BCDCOG explained that their population forecast procedure first generates the entire population, then splits this population into households and group quarters, and finally generates individual households by household size. The household size decreases in forecast years to represent the aging population and reflect recent trends. A panelist emphasized that representing the aging population is a pressing need for MPOs because of the change in percentage of work-based travel, mode share, and travel time of day. Another panelist encouraged BCDCOG to consider modeling age explicitly and shared the following via email:
Many years ago, I was brought in to help on a model where the population was declining, yet volume was growing at an annual rate of 2%. The issue was, the number of children and people without access to autos was declining. Yet, older people as a population group was growing. Also, auto registration was also growing, and growing at a rate faster than the population group growth. Thus, those who could drive, had greater access to autos, and thus the traffic growth. Basically, the area was transitioning to a retirement community.
BCDCOG explained that household income is represented by proxy through the number of workers in the household and vehicle ownership. BCDCOG acknowledged that there is some difficulty in distinguishing low-income zero-worker households from retired person zero-income households. Several panel members expressed concern that this approach is insufficient to capture the household economic situation for two major reasons. First, retired persons would not be classified as working yet may have a high amount of travel. Second, affluent households are trending towards zero-auto ownership, but this does not constrain their ability to travel.
The panel recommended that BCDCOG include age and income in forecasts to account for the growing retiree segment of the population. The panel discussion around vehicle ownership is recorded in section 1.2.
BCDCOG explained that their economic forecasts are based on data from InfoUSA. BCDCOG checks InfoUSA data against the US Bureau of Economic Analysis and large local employers. But the overall employment rates are largely driven by population data, in which there is more confidence and the projected employment growth rate is essentially constrained to match the population growth rate. A panelist pointed out that the increasing retirement population will require a different mix of employment than the historic population (e.g. retail, restaurants, health clinics) and that trip-making behavior for retirees is markedly different. This panelist explained that the new development areas, which are likely to be predominantly retirement communities, should not be assumed to have a similar mix of population and employment as existing areas.
Several panelists questioned the use of an essentially fixed relationship between employment and population. There are many factors that will influence the region's economy such as the port, large employers, and development activity. One panelist pointed out that the port is more dependent on macro economic forecasts than the local population. Another panelist noted that there have been recessions in the area during recent history so it is not necessarily reasonable to forecast direct growth. Several panelists recommended a scenario planning approach to capture the range of potential future conditions. BCDCOG has prepared four different land use scenarios for planning, although the outcome is primarily variations in the number of households because employment is balanced to households in distribution.
REMI economic forecasts had been used in the past by BCDCOG to produce economic forecasts. A panelist explained that their agency uses REMI for control totals and a similar approach could be implemented here.
The panel recommended that employment in the region not be assumed to be strictly correlated to population, but rather should follow other economic indicators. Economic forecasts should be developed with input from state and local economic forecasters and local commerce representatives. Moreover, the employment forecasts should be made through multiple scenarios and not as a single value.
BCDCOG showed that transit drive and walk access links are included in the master network. This is convenient to ensure that changes are made in only one place but the links are virtual and can get in the way of using the network. A panelist agreed that this is a cumbersome way to formulate the network and that the virtual links should be stored separately or generated on demand.
There are no network restrictions on truck assignment although only the historic downtown is prohibited in reality. Intersection capacities, turning movements and other restrictions are not represented in the network. Instead this is represented through capacity that varies by functional class and area type.
The panel agreed that the network should be improved to include more information about intersection delays, turning movement restrictions and special allowances. Without this type of information, many transportation system management (TSM) strategies cannot be evaluated. Furthermore, these features are necessary if the network were to be used for microsimulation.
Household cross-classifications are by number of workers, household size, and vehicle ownership. Trips are generated for four purposes: Home Based Work, Home Based School, Home Based Other, and Non-Home Based.
BCDCOG employs five area type profiles (CBD, Urban, Suburban, Rural, and Beach) to disaggregate households by size, worker, and vehicle ownership. A panelist noted that the Beach communities are predominantly retirement communities and thus have their own characteristics. Retirement communities that are not on the shore, however, would not be categorized as "Beach" area type and thus would not be as well represented. Another panelist recommended that BCDCOG evaluate the seasonality of Beach community populations and potentially work with states that have similar seasonal variations, such as Florida.
BCDCOG explained that the household segments categorized vehicle ownership as a binary state (i.e., whether or not the household owns at least one vehicle). Also, vehicle ownership is not influenced by accessibility or other zonal characteristics beyond the area type. Several panelists emphasized that a vehicle sufficiency model that models the number of vehicles per worker or per driver is more appropriate to truly represent captive transit riders. Moreover, a vehicle availability model should be sensitive to the transit accessibility, particularly given the trend towards lower vehicle ownership in affluent, dense, well-connected areas.
A panelist suggested that the trip generation model include segmentation by age to capture not only the unique travel behaviors of retirees, but also the influence of children on household trip making. This panelist further recommended that BCDCOG consider a person-level trip generation model stratified by working adult, retirees, other non-working adults, and children. In this case, trip rates would be based on the household and person type nexus.
In addition to the recommendations that retirement communities and vehicle availability be better represented in the model, the panel recommended that the non-home-based trip purpose should be divided into two purposes: work and other. The non-home-based trip purpose segmentation should then be leveraged in both the distribution and mode choice models. The home-based-other trip purpose should be divided into three purposes: home-based-recreation, shop, and other.
As described in Section 3.2, the CHATS model includes special generators for multiple types of trips. Several panelists pointed out that special generators are not desirable in a model because of their insensitivity to other conditions. Special generators are effective to improve the fit of the base year model, but have poor forecasting power.
One panelist explained that they were able to eliminate special generators entirely from their model using a procedure to refine discretionary trips by industry. BCDCOG noted that this requires a higher level of survey data than is currently available. Another panelist explained that trips currently represented through special generators can be more easily incorporated when a destination choice model with the correct specification of attractiveness (right employment by industry) and the right trip matrix balance procedure (production and attraction trips) is implemented in the model.
Another panelist discussed that the military base is unique and operates as a "black box" so this would need to be modeled using a special generator but that all other special generators should be incorporated into the model.
In regards to externals, a panelist recommended that the external trips be generated as person trips with work and non-work purpose segmentation. This approach would make the external trip mode sensitive to network conditions.
The panel recommended that BCDCOG work to eliminate all special generators from their model with the exception of military, which is infeasible to represent otherwise.
BCDCOG requested advice from the panel on representing visitor activity or seasonal variations as there is no current functionality in the model to do so. An ongoing study at the College of Charleston is collecting updated hotel data as well as visitor trip making and auto propensities.
The panel agreed that the lack of a visitor model is a significant shortcoming of the CHATS model. BCDCOG mentioned that the tourist peak lasts 8-10 months a year therefore a typical work day model includes some visitor travel impact. The visitor levels fluctuate over the course of the season so, in addition to the ongoing hotel survey, the panel suggested that BCDCOG conduct a visitor survey. A panelist pointed out that the non-motorized travel mode is critical to support the visitor module as many tourists do not use autos for trips in the downtown region.
A panelist cited the implementation example done in Knoxville to include travel to the Great Smoky Mountain National Park, which is not actually contained in the Knoxville TPO Planning Area but that effects of this park are evident. The Knoxville model includes a capability to adjust hotel/rental unit occupancy levels to test demand at different times of the season.
Another panelist reminded BCDCOG that the regional model is not responsible for capturing special events although regular season variation should be represented.
The panel recommended that BCDCOG collect more data on visitors in the area and implement a visitor model similar to the Knoxville model.
BCDCOG identified the current gravity model as one of the primary areas of concern in the model. The gravity model does not adequately account for the psychological barrier that bridges present and simulates too many trips between peninsulas. K-factors are necessary in the person trip distribution model to keep trips within each peninsula and manage the bridge volume in the model. The distribution model uses a time-of-day weighted highway time, not a composite highway and transit impedance factor.
During a conference call held with the panel prior to the peer review meeting, a panelist suggested a tri-proportional gravity model. This type of model is in use at the Boston MPO and has proven capable of representing the deterrence of travel bridging geographic barriers, such as rivers and bays.
During the meeting at least two panelists recommended implementing a destination choice model. One panelist had estimated a destination choice model for an area with similar bridge crossings and even state crossings with different sales taxes. Furthermore, the destination choice model can use composite logsum formulations from the mode choice so that the trip distribution is sensitive to both auto and transit accessibilities. A panelist suggested that BCDCOG could transfer the model from an area such as Portland, OR and recalibrate to fit their trip length distribution. Another panelist also proposed conducting destination choice at the person, rather than household level. Given that BCDCOG is already using CommunityViz, the person-level forecasts would not be overly difficult to produce.
Panel members agreed that it would not be a difficult upgrade to improve the distribution model and that this should be moved from a long-term to a short-term project. The panel concluded that a destination choice model which can include travelers' characteristics, transportation system characteristics (incl. impedance), and destination zone attributes should be implemented.
BCDCOG explained that the previous mode split function included a share for non-motorized modes, but that the current mode choice model does not. Several panelists raised concerns about this because the trip generation rate was estimated using all household trips including non-motorized trips. Therefore, the trips processed in the mode choice model are for all modes while the current mode choice model is constructed to only produce motorized trips. BCDCOG estimates that at most 5% of trips are non-motorized and that many of these would be intrazonal. Yet, a panelist pointed out that this is a higher share than transit riders and that the historic downtown has a higher concentration of non-motorized travel. BCDCOG agreed that the motorized trips are being overstated in the model, particularly for short distances. BCDCOG explained that there have been some studies into the quality of non-motorized trails and byways, but that the data is dated. A panelist suggested that a sketch planning tool could be a good first step to developing an integrated non-motorized model. Another panelist explained that a non-motorized split could be implemented prior to the motorized mode choice model or, alternatively, the current mode choice model structure could be extended to include non-motorized modes, which would be more difficult.
A panelist recommended that the region set a mode shift goal and track a mode share metric as part of the model results. BCDCOG had identified a mode share goal in the previous LRTP and is doing a fixed rail study now, but has not yet included mode share in the model metrics.
BCDCOG explained that there is a fixed operating cost parameter in the auto mode utilities, but that parking cost data is not available. Two panelists pointed out that the only socio-economic input to mode choice is vehicle availability, which is insufficient particularly because vehicle availability is not well incorporated in the model.
Transit ridership in the area has doubled over the last 10 years and it is unclear if the model is representing current transit ridership accurately. A panelist pointed out that transit ridership can be strongly driven by college students, but that the college trips are not segmented in the model. BCDCOG noted that a majority of transit routes serves the corridor into the College of Charleston. The panel suggested that segmenting Home-Based College trips may be necessary to more accurately represent the strong connection between transit ridership and college trips.
The panel recommended that non-motorized trips be explicitly handled within the CHATS model, either by extending the mode choice model to incorporate non-motorized alternatives or adding a non-motorized mode split model prior to the current motorized mode choice model. The model output must show that non-motorized travel is accounted for in the modeling process, although the non-motorized travel may not be assigned. In general, the motorized mode choice model should include more socio-economic variables across all alternatives.
The CHATS model uses a standard BPR curve with parameters from NCHRP Report 365 (page 95, table 48). Minor adjustments have been made based on observed congestion. There is no plan to update these curves as BCDCOG's understanding of congestion is due to bottlenecks and weaving and not capacity constraints.
A panelist suggested that BCDCOG evaluate the traffic levels during the midday time period and consider separating the midday period from night, which are currently assigned together, to capture the midday rush.
The model has not been validated using speed data, and so adjustments to the BPR curve could not be made with confidence.
The panel recommended that BCDCOG revisit the BPR curves on their major corridors once speed data is available.
BCDCOG validates the model according to daily traffic count data, segmented by facility type, examines the overall model RMSE, and compares the county-level VMT to estimates from the South Carolina Department of Transportation (SCDOT). The 2003 household survey results are used to validate trip-length distributions and district-district flows.
A panelist pointed out that the count data is in AADT format, but the model is producing AWDT. Another panelist seconded this and explained that they observed other instances in the model documentation that indicated issues with the model input data that need to be corrected.
A panel member commented that the current validation process is insufficient and that more attention needs to be paid to each component in the model process and not only the end assignment results. Another panel member agreed and stressed that speeds need to be included in the validation metrics. There is a lot more variation in speed than volume in a model, particularly when links are near capacity.
Currently there is no speed data used in model validation, but BCDCOG will have access to INRIX data through their membership in the I-95 corridor coalition. The data received will be in 5 minute intervals on the I-26 and I-526 corridors. A panelist familiar with using type of data from INRIX, NAVTEQ and HERE cautioned BCDCOG that the data is free but extracting useful information can be time consuming and requires a dedicated analyst with a strong skill set.
BCDCOG highlighted that the existing truck model has not been updated since 2003 due to a lack of more recent classified counts, that the Quick Response Freight Manual (QRFM) distribution requires K-factors to meet the observed traffic, and that distribution centers are not being properly modeled. BCDCOG is most concerned that freight activity related to the Port of Charleston is not well represented. It has been difficult for BCDCOG to get data from the port, and the information they are using is limited to vehicle counts at the access points. An updated freight model has been identified in the UPWP.
A panelist explained that there are other ways to gather information about port operations and freight activity in the region. This panelist confirmed with the SCDOT representative that they have funded MPO access to TRANSEARCH data at the county level, which can be very useful because the entire port is within Charleston County. The panelist recommended that BCDCOG obtain a copy of the 5 year plan from the port to determine what types of commodities will be handled over the near future. This will indicate the types of warehouses and distribution centers that will be constructed in the region. The panelist also recommended that BCDCOG obtain a copy of the bill of lading information to understand the types of commodities that are currently coming into the port. The growth of Berkeley and Dorchester counties, where the warehouses and distribution centers are expected to be located, could be very different depending on the types of commodities coming into the port. Another panelist pointed out that the port-based trips would be traveling out of the region, therefore the existing QRFM-based truck model component should remain in the model.
The existing BCDCOG model is a truck model and not a true freight model in that it does not account for goods movements or commodity flows. A panelist explained that freight movements are driven by costs, not distances. A discussion developed around the different types of urban goods movements and how varied the delivery frequency can be based on the type of product and technology. The panel agreed that complexities of freight travel imply that they do not necessarily need to be integrated into the model. Moreover, when freight forecasts of more than 10 years are included in the model they need to be treated as highly speculative and should be done with a scenario planning approach.
To build a freight model, however, a larger geographical context is necessary. The representative from SCDOT expressed interest and support to a larger freight model. A panelist explained that the statewide freight model flows could serve as an input to the local model and that there are methods to disaggregate to a zonal level. The group also noted that South Carolina includes two Class 1 railroads and that the statewide intermodal plan created a freight committee. Panelists recommended that this committee be included in any freight model formulations.
Several panelists agreed that BCDCOG needs to more closely identify the specific need for a freight model in their region, beyond the presence of the port and large employers. By balancing employment inputs to population, BCDCOG is undermining the influence commercial forces have on the model anyway. A panelist noted that a freight model is not necessary to build a freight plan for the region. A freight plan could be based on a data-driven process leveraging the TRANSEARCH data already available from SCDOT.
The panel recommended strongly that BCDCOG not pursue a tour-based freight model to integrate into their regional travel demand model, but that they should be tactical depending on the identified need for freight information to the degree that a "freight model" need not even be constructed. That is to say that the urban goods component does not need to be fully integrated into the regional travel demand model; instead, it could operate as a data-driven exogenous component that provides inputs to the travel demand model. The panel also recommended that BCDCOG include SCDOT and the port authority in future efforts to build a freight forecasting component.
BCDCOG are the primary users of the CHATS model. There are 27 municipalities represented in the model area, and some municipalities have requested access to the model for land-use studies although not for transportation-related analysis. The model is also used by consultants for traffic impact studies and their feedback has been neutral with only minor network errors found.
SCDOT has incorporated the CHATS model into the recently completed statewide model. The statewide model assembled the regional MPO networks and conducts a geographically constrained distribution of HBW, HBO, and NHB trip purposes. The geographic constraint assures that the statewide model volumes are consistent with the MPO network externals. Most important, a freight network is currently being coded into the statewide model. Once complete, this information will be available to the MPOs.
SCDOT reported that their evaluation of the CHATS model forecasts is that the model under predicts volume on the arterials compared to the growth trend based on count data. However, the forecast year model network only includes Existing and Committed projects, which implies that the arterials may be at capacity and would see a continued growth in volume if capacity were increased. A panelist suggested that an improved traffic count forecast could be developed using local knowledge and land-use growth patterns to determine if the model is actually under-predicting.
A panel member asked if there are any tolled or public-private-partnership projects in the planning horizon. BCDCOG explained that there are tolled facilities in South Carolina, but none in the planning region nor any in the planned projects, however, there is concern at the state level as with almost every state with how to fund transportation projects. The panelist explained that Public-Private-Partnership (PPP) projects require a higher level of rigor in the model and that a better handling of household income, for example, would be necessary.
The model has been developed with a series of custom interfaces to select different build scenarios.
A panel member recommended that BCDCOG develop their future models with a more generic TransCAD interface. This will make the model easier for the average TransCAD user to understand, which will help in hiring new modelers and encouraging other users. Also, maintaining a generic format will make the model easier to upgrade to future versions of TransCAD.
BCDCOG is opting for 1,250 supplemental NHTS samples, 1,000 of which will be within the CHATS model area. In response to BCDCOG's question about the upcoming NHTS add-on, a panel member recommended that BCDCOG specify that only weekdays be sampled. Otherwise, the survey will distribute equally across all days and BCDCOG may only be interested in the weekday travel. Panel members also recommended that any additional questions focus on the areas of interest for BCDCOG, such as to support a new destination choice model. The panel recommended against transit questions because area ridership is very low. Transit-specific data should be collected through an on-board survey.
Two panelists pointed out that in order to make use of vehicle probe data to generate speeds for model validation BCDCOG needs to add capacity to process the data in-house by hiring an analyst with data mining experience. Moreover, this additional analyst would need to have an active peer group to learn and consult with on data applications. The panel recommends that BCDCOG look to SCDOT for leadership on using this type of data as well as DOTs and MPOs from neighboring states.
A panel member noted that it is easier to implement TransModeler from a TransCAD regional model starting point as opposed to other micro/mesosimulation packages, but only relatively so. Micro/meso models require detailed information on turning lanes and signal timing, which is not currently included in the existing CHATS model network.
