Virginia Department of Transportation (VDOT) Travel Model Improvement Program Peer Review

5.0 Peer Review Panel Recommendations

In the last half of the meeting, the peer review panel took about one and one half hours in an executive session, closed to all participants of the meeting except for the panel members. The reason for this was to allow panel members to speak freely and openly among themselves while developing formal recommendations. This section details those recommendations of the panel.

5.1 Metrics

The panel recommended that VDOT get very specific (and soon) about the measures of effectiveness and performance measures that the model will generate. More specifically, the panel suggested that VDOT consider the following questions/issues:

• What will the model generate and what are VDOT's needs from the model?
• How will the model results be presented (e.g., percent shift under scenarios or actual forecasts)?
• Clearly identify the role of this model in relation to various HB 2 performance areas. What will it estimate, and how will those results be used? Evaluate whether the developed model meets those needs.
• Do not use the model for conformity analysis, but it might help as a planning tool with environmental performance indicators from post-processing.
• There is no "one model that rules them all"; the statewide model is just one tool in the toolbox.

Most importantly, the panel emphasized that VDOT always keep an eye toward their needs. Keep expectations reasonable and remember that there are things a statewide model will not be well-suited for (such as modeling intra-urban travel). Keeping these issues in mind will be useful.

5.2 Price/Tolling Sensitivity

The panel asserted that price sensitivity and tolling sensitivity are important, and they should be accommodated in the model structure. Additional specific recommendations are as follows:

• These sensitivities can be handled through asserted values of time by trip purpose. A detailed local study would be required to do anything else (other than assert the values). It would be good to store values of time in an input file so that alternatives could be adjusted and tested.
• Look to local MPOs that may have done value of time studies for guidance.
• Using a before and after study may be a good way to do the estimation of values of time. Other methods are also plausible.
• Value of time is likely to be stable for trucks. It would take a study to determine difference in response for trucks in urban and rural areas.

The panel emphasized that VDOT test different values of time once the model is up and running. It is critical to look at how value of time affects results and how different "reasonable" values of time might generate different outcomes. Value of time is one area of uncertainty in the model. The literature can offer a range of plausible values of time, but sensitivity testing will help determine where in that range VDOT should be for their state.

In addition, the panel recommended that VDOT allow for significant shifts in vehicle occupancy that could affect toll costs and price sensitivity. This should be a feature built into the model software. The panel did not think that VDOT would necessarily use it or need it, but it will be good to plan for these sorts of possibilities.

5.3 Model Structure

VDOT should pursue development of a four-step, trip-based model, which is what VDOT planned for. However, the panel recommended that VDOT consider only minimal mode choice elements, rather than the more elaborate mode choice models VDOT planned. Additional specific recommendations are as follows:

• The model should be a person trip model, with simple factoring to vehicle trips.
• VDOT may consider a separate model to develop nuanced person-to-vehicle factors.
• The panel recommends deferring all transit considerations until a later phase.
• However, VDOT may want to consider intercity bus and/or rail at this stage.
• The panel recommends deferring aviation to a later phase as well (though the model may still consider airports as special generators).

The panel also suggested using a structured repeatable process for developing the required model parameters. A user guide that takes the owner through specific steps for developing model parameters from the data would be useful. Another reason for doing this is that it can be important for transparency with outsiders, and it can also be important if legal issues related to model forecasts ever arise.

5.4 Truck Model

The truck model should be a commodity-based model. In addition, the following recommendations/comments were made with respect to a truck model.

• Commodity-based models require a lot of data (e.g., FAF, TRANSEARCH).
• Keeping the model current will require an on-going effort.
• There should be a disaggregation step from counties to TAZs.
• Establishment information and LED may help disaggregate commodity data.
• The truck model should look at possible future shifts from truck to rail, or look at different growth rates in different commodity classes.
• VDOT should include FAF in truck data sources.
• VDOT should consider using ATRI and other origin-destination truck data to help disaggregate the commodity data.
• The panel suggested that there may be an opportunity to work with TRANSEARCH to develop estimates that match the state's truck counts. This has been done with success in other states. In this case, the panel recommends splitting the work into smaller pieces (e.g., three or four) and having multiple review points.

With respect to rail freight movements, the panel recommended that VDOT consider the Rail Carload Waybill sample data as an additional data source.

VDOT had proposed to use ATRI-based data for truck modeling, which is not commodity-based. One panel member remarked that ATRI data does not have any policy sensitivity in it, and a lot of corrections would be needed to use it well. Another panel member noted that the types of policies VDOT may be interested in are mode shifts between truck and rail and shifts across commodities. These questions cannot be answered with ATRI data, but can be with commodity flow data. However, the panel noted that ATRI and other disaggregate data sources can still be useful, and, for instance, can help disaggregate commodity flows from county level to TAZ level.

Another panelist noted that given the short time frame VDOT is working under, there really is not enough time to develop a rail network. Unfortunately, that is really what is needed to do any sort of mode split for freight. However, it would be possible to do some types of elasticity analyses to drive the truck component of freight, making it sensitive, in some ways, to modal type issues.

The panel also recommends that VDOT consider smaller trucks in their model, which are not represented in TRANSEARCH data.

• Smaller trucks tend to be shorter, more frequent trips.
• TRANSEARCH data does not provide this.
• The panel recommends that VDOT look to the commercial vehicle survey conducted in Calgary (and possibly in Ohio) as places to start.
• In the near term, the model structure should be designed to allow for representation of these trucks in the future (including placeholder data). This will allow for a seamless transition when local data for these trucks becomes available.

5.5 Model Geography and Network

The panel made several recommendations related to the extent of the modeling geography, as detailed below:

• If not already doing so, the panel recommends that VDOT aggregate MPO TAZs so the MPO demographic data can be easily incorporated, and consistency with MPO models is achieved.
• The panel recommends a 48-state model with a minimum number of external stations for both freight and passenger travel. For external passenger interstate travel, cell phone data (such as AirSage) could be useful to generate origin-destination trip tables, but it may be expensive.
• VDOT should consider using the LEHD to study trips to/from the halo. The halo refers to areas external to the state, but where modeling geography is still relatively disaggregate. These areas are close to the state boundaries.
• VDOT should consider setting external stations farther from the state boundary, to allow for shifts in state entry points in response to network changes.
• Additional recommendations were made with respect to developing networks for the model:
• The panel recommends retaining as much detail from the base HERE network as possible (e.g., freeway ramps and interchanges).
• Minor roads can be removed, like locals and minor collectors.
• Procedures should be developed for attaching auxiliary information to the network links (e.g., functional class).
• Future year networks get harder to maintain with greater detail.

5.6 Consistency with MPO Models

The panel recommended that VDOT concentrate on validation outside the major MPOs, but attempt to reproduce all traffic with consistent trip generation and trip distribution (which would not necessarily be the same as the MPO model). In addition, the panel made the following points:

• Key validation points for counts are external stations to the major MPOs.
• It may be useful to look at through corridors in urban areas (long distance corridors). In such cases, enough detail in the links and zone structure will be needed to ensure shifts in traffic remain plausible.
• As zones are enlarged, network detail can shrink (with more trips becoming "intrazonal"). There is a balance here that should be considered.
• Differences in zonal and network detail between the statewide model and MPO models deserve attention to ensure the statewide model's network links in and around major MPOs are assigned reasonable traffic volumes (which result in reasonable speeds on those links).
• The panel recommends that VDOT use as much detail as can be managed with reasonable model run times.

5.7 Time-of-Day and Traffic Assignment

The panel recommended that time-of-day modeling should ideally be done in one-hour increments, with suitable disaggregation of trip tables. For instance, VDOT could develop a 24-hour demand table, factor to single hours, and then assign. However, this may be beyond the scope of phase 1 model implementation. For phase 1, aggregating hourly demand to four time periods and using static assignment procedures would be reasonable.

One panel member emphasized that time-of-day is critical, and it is not possible to do any sort of air quality or congestion analysis with 24-hour assignment. There are some biases with static assignment over one-hour periods, since some biases can emerge, particularly for a statewide model with long trips that may not complete within one hour. However, 24-hour assignment loses a lot of information that the model could be generating.

The panel recommended that assignment be capacity constrained.

• It could be multi-class assignment, or it might preload trucks to the network for assignment.
• In addition, VDOT should expect to pay close attention to speeds, and ideally, junction details to support assignment.

One panelist remarked that it is not possible to model intersection delay using volume-delay functions, as VDOT proposed. The 2010 Highway Capacity Manual (HCM) has tools for estimating delay for roads with intersection delay. The panelist contends that speeds are an important output to transportation models, and that can only be reliably performed if intersections are modeled correctly. Unfortunately, coding intersections (and their delay) correctly can require a great deal of time (e.g., 30 minutes per intersection), which probably eliminates this option for VDOT at this stage, but should be considered in later phases of model development if possible.

The panel believes that dynamic traffic assignment (DTA) is an option that VDOT should consider. However, this could be added to the model at a later time. Some specific recommendations/comments related to DTA are as follows:

• Without DTA, longer trips may not match well spatially.
• DTA may not require much higher level detail than VDOT is already planning.
• DTA may have trouble dealing with trips longer than a day (but the analyst can compensate with longer assignment periods, e.g., 72 hours).
• While the panel believes DTA might ultimately prove the best option, they suggested an acceptable baseline would be static equilibrium assignment with four time periods, using multi-class assignment procedures.

5.8 Other Items

The panel made several additional recommendations on a variety of topics. These recommendations are outlined below:

• The panel encourages VDOT to develop a long-term statewide data clearinghouse. This would be a standard place where all data is stored. In this way, all modelers in the state would have an inventory of all the available data and everyone would be starting with the same datasets.
• Developing a working model (with appropriate sensitivities) in the short time frame stated by VDOT will be very challenging. The panel recommends that VDOT keep the statewide model off of the short-term critical path if possible.
• VDOT should strive to key payments to their model development contractor off specific milestones and validation standards, with a clear definition of what is acceptable.
• VDOT should sync the base year with the 2010 Census. For future years, the panel recommends pivoting from the base year.

A couple of panel members offered some concluding remarks. One panelist suggested that developing a longer range plan for model development can be important (e.g., 3 and 5 years). In addition, it is important to keep in mind that there is a suite of models that can be used for analyses. The statewide model need not be capable of doing all analyses on its own.

Another panel member commented that the biggest issue will be development of person trip tables (which is an issue everywhere, and not just for statewide models). The panelist fears that even after VDOT goes through the process of destination choice and mode choice model development, the accuracy of the resulting trip tables may be quite poor. While a couple of other panel members believe that AirSage data could be a fresh source of origin-destination data that could be useful to verify model results, a different panelist does not think AirSage data is all that accurate either, and has encountered multiple agencies using origin-destination matrix estimation on AirSage data in order to validate the model.