Model Setup and Run

This section describes the model configuration and run options with recommendations by analysis type. The areas are organized by:

Full run vs. assignment-only

Both the set of activity based demand components (CT-RAMP) and the network assignment (highway and transit) procedures can have long run times. When the feedback process between these two sets of procedures is run, this requires multiple iterations of both of them. It is therefore useful to determine for which types of model runs a simplified application process might be sufficient to produce the necessary outputs.

There are three basic ways in which the model may be run (listed from most complex to simplest):

1. Full model run including all demand and assignment components with full feedback;

2. Model run without feedback, in which both the demand and assignment components are run, but only for a single iteration without feedback; and

3. Assignment only, where the outputs from the demand components are obtained from a previous model run (“seed” scenario) and only the assignment procedures are run. If the scenario to be analyzed requires only highway assignment outputs, or transit assignment outputs are not expected to significantly change from the seed scenario, it will be possible to run only the highway assignment and to skip the transit assignment.

Note that the feedback process is used to improve consistency between the highway travel times used as inputs to the demand models and the travel times realized from the highway assignment process. Due to congestion, the highway travel times are dependent on the assigned volumes. In many cases, the input travel times are already close to the times from the assignment because the model scenario is based on one that produces similar highway assignment results. In such cases, the feedback process is not necessary.

The choice of the model application process depends on the type of planning analysis scenario to be run. Usually, a scenario is based on a previously run “base” scenario—for example, a scenario for a highway project may be developed from a “no build” scenario. Each scenario requires specific changes to the model inputs used in the base scenario (and in some cases, model parameters). The major model inputs are the transportation system (represented by the highway and transit networks) and the land use/socioeconomic assumptions (represented by the socioeconomic data files). Other input data changes might include policy related changes that are not represented directly by the major inputs (for example, parking cost or auto operating cost changes). The more substantial the changes in model inputs (in terms of magnitude and geographic range), the more complete the application process that is required.

The following list specifies the types of model runs that are appropriate for each type of application described in the introductory section.

1. Long range planning scenarios – These scenarios always require full model runs with feedback as they include new sets of land use/socioeconomic data (often for a new forecast year) and changes to transportation networks, which are usually substantial.
2. Analysis of specific highway improvement projects
   1. Major – Since highway capacity changes are substantial, a full model run with feedback is always required.
   2. Minor – Smaller projects that do not significantly increase roadway capacity should not change congestion levels, and therefore feedback is unnecessary. It is likely that these projects would not affect choices of how much, where, when, and on which modes to travel very much, and so it is likely that assignment only is sufficient (in this case, highway assignment only).
3. Analysis of specific transit improvement projects
   1. Major – These generally require full model runs, since travel patterns beyond mode choice may be affected. In most cases, shifts in transit use patterns would not substantially affect congestion, and so model runs without feedback would be sufficient.
   2. Minor – These changes should usually be minor enough that transit travel patterns can be assumed not to change, and so assignment only is sufficient (in this case, transit assignment only).
4. Regional land use development or demographic changes – These scenarios always require full model runs with feedback as they include new sets of land use/socioeconomic data.
5. Localized (site specific) land use changes
   1. Major – Since these developments may have significant impacts on travel patterns, a full mode run will be needed. These could be full model runs with feedback if the developments may affect congestion on major facilities; if not, model runs without feedback may be sufficient.
   2. Minor – These scenarios should be minor enough that assignment only is sufficient. If the development is not located near transit service, transit assignment may not be needed.
6. Roadway pricing/managed lane projects – Unless these projects represent only minor pricing changes (e.g., a modest toll change on a single facility), it is likely that mode and perhaps destination choices will be affected by the pricing policies. It is probable that, unless the toll amounts are substantial or roadway capacities are significantly affected, feedback will not be required, and therefore model run without feedback will be sufficient.
7. Active transportation/non-motorized travel projects – A very minor project might not require a model run at all since non-motorized trips are not assigned in SERPM 8. For projects that might have an effect on mode choice, model runs will be needed. Since non-motorized travel represents predominantly short trips, substantial changes in VMT and congestion are unlikely, and so model runs without feedback should be sufficient.
8. Environmental justice analysis – In this type of analysis, the effects of various policies or projects on specific segments of the population are required. Since trip assignment is not done by population segment, the demand model components must be run. Whether feedback is required depends on the scale of the policy or project to be analyzed. If they require environmental justice analysis, they probably will usually be major in nature, and then a full model run with feedback will be required.
9. Analysis of emerging technologies (e.g., autonomous vehicles, TNCs) – Of course, there is much more uncertainty involved in analyzing emerging technologies on which there is little or no data available for model estimation or validation. However, whatever the assumptions, the behavioral changes are expected to be so substantial that full model runs with feedback will be required.

PopSyn

PopSyn generates a synthetic population for the model region. The synthetic population is one of the main inputs to the demand model, which simulates the activity pattern and travel for each person in the population.

PopSyn generates the synthetic population using control totals for various person and household characteristics (as well as group quarters population) and a seed table based on data from the American Community Survey (ACS). The control totals are the following:

Household Controls:

• Housing Unit Type
• Household Size
• Household Income
• Number of Workers in Household
• Presence of Children in Household

Person Controls:

• Age
• Occupation
• Gender
• Race/Ethnicity
• Type of Group Quarters

Since the seed table comes from the ACS data and is unaffected by the assumptions of a particular application scenario, PopSyn needs to be run only if the control totals are changed from the base scenario. This will be the case for a new forecast year scenario, or for a scenario where the number of persons or households in any TAZ or group of TAZs is changed (for example, a scenario that assumes a different locational distribution of population growth), or when any of the characteristics of the population that are related to the controls is changed (for example, if the income distribution of households is changed).

Note that this means that PopSyn need not be run for a scenario where the only changes from the base scenario are related to the highway or transit networks. If it is assumed that network changes are significant enough that the underlying land use would change, then these changes would already be reflected by revisions to the population assumptions (and therefore the control totals).

Shadow pricing

Shadow pricing is a process within the Usual Workplace and School Location models to ensure consistency at the MAZ level between simulated workplace locations and employment totals and simulated school locations and students respectively. Shadow prices are derived through an iterative process during usual work and school location components and set prices by worker occupational segment (White Collar, Services, Health, Retail and Food, Blue Collar, and Military) and by school type (Preschool, Grades K-8, Grades 9-12, University and by District [County]).

Shadow prices have been developed for the base 2015 and forecast 2045 scenarios. The shadow prices are defined by two scenario keys (one for work, one for school) and are part of the Land Use Inputs section of the scenario manager. Changes in land use (employment totals and type, school enrollment, and population) may require a different shadow price to maintain the consistency between usual work and school locations and the jobs and enrollment.

When to generate new shadow prices?

Although shadow prices are sensitive to changes in land use and the network conditions, it is not recommended to generate new shadow prices for every new scenario. Testing with substantial network changes, but no land use changes, showed that the worker-job and student-enrollment consistency was generally maintained. Moreover, a change in the shadow price input will create differences in the model outputs and may complicate comparisons between scenarios. In addition, the shadow pricing process is time consuming and requires about as much time as one speed feedback model iteration (~4 hours).

Note that in cases where assignment only applications are appropriate, the demand models (including Usual Workplace and School Location) are not run, and therefore shadow pricing is not used.

Shadow prices should be generated when there is a “substantial” change in the land use inputs. The definition of “substantial” depends on the previous land use and existing shadow prices. When in doubt, the modeler can refer to the shadow price summary report (_reports\workschool_consistency_ScenarioName.xlsx). This report presents a super-district, county, and regional summary of the average and maximum ratio of worker/student to job/enrollment as well as an MAZ-level scatter plot. As a reference, a copy of the 2015 and 2045 reports are available in the \Inputs\SEData\shadow_prices folder of the model catalog.

How to generate new shadow prices?

The ShadowPricing scenario in the model scenario directory has been configured to automatically generate an updated set of shadow prices. The following keys MUST be configured (in priority order):

Land Use Inputs

• MAZ Data (employment, enrollment data at the MAZ level)
• Persons File (synthetic population person table)
• Households File (synthetic population household table)

Note that the shadow price file inputs are not used and can be ignored.

Run Controls

• Loaded network to see transit skims
• AM Seed Skim
• PM Seed Skim
• Off-Peak Skim

The loaded network and seed skims must be consistent with the forecast year of the land use data. For example, if the land use data scenario is generally consistent with the 2045 forecast, the 2045 loaded network and seed skims should be used.

Networks

• Highway network
• Transit network

It is not necessary to set any other model inputs, including the Special Generator Inputs. The ShadowPricing scenario will only run through the ABM components before exiting; therefore, assignment inputs are not needed.

The ShadowPricing scenario can be launched in the same manner as any other scenario (e.g. by pushing the Run button on the scenario manager). Cube will run through the initialization steps and launch CT-RAMP to generate the new shadow prices. The scenario is configured to exit after the shadow prices are complete with a benign error. The modeler can confirm the process was successful by checking the \Scenarios\ShadowPricing\_abm\ folder for the following files:

• ShadowPricing_new_work_5.csv
• ShadowPricing_new_school_5.csv

These two files need to be renamed with a descriptive term specific to the scenario and copied to the \Inputs\SEData\SEData\shadow_prices folder. These files should then be pointed to in the scenario manager shadow prices keys under Land Use Inputs.

Summary

In summary, when there is a question about if new shadow prices are necessary, the following steps to evaluate and generate new shadow prices are:

1. Run the model with the updated land use inputs
2. Review the workschool_consistency_ScenarioName.xlsx report and check for inconsistencies.
3. If inconsistencies outside of the standard range are present:
   1. Configure the ShadowPrice scenario with the updated land use inputs, forecast year consistent seed network and skims, and highway inputs.
   2. Run the ShadowPrice scenario and confirm creation of the new school and work shadow price files
   3. Rename and copy the files to the Inputs\SEData\shadow_prices folder
   4. Update the test scenario definition to utilize the new shadow price files and rerun.

Seed skims

The “seed skims” refer to the matrices of network level of service variables, especially highway travel times, that are used as inputs to the initial iteration of the demand component execution for a new scenario. They are therefore applicable to a full model run with feedback or a model run without feedback. They are not used in an assignment only application since the demand components are not run.

In a model run with feedback, the choice of what to use for the seed skims can significantly affect the model run time, the level of convergence of the feedback process (i.e., how close the output travel times are to the inputs), or both. If a set number of feedback loops is enforced, then the run time is not affected much by the seed skims, but the level of convergence is highly dependent on how close the seed skims are to the eventual outputs. If the feedback process is run to a prespecified convergence level, then the number of loops required to converge, and therefore the model run time, depends greatly on how close the seed skims are to the eventual outputs.

It is obvious that choosing seed skims as close as possible to the expected outputs is advantageous. This choice is of course dependent on what sets of skims are available to serve as the seed. Although they are always available, free flow skims are not an advantageous choice since free flow conditions do not reflect the substantial traffic congestion that is present in Southeast Florida. Therefore, the choice should come from the set of previous model runs whose results have been validated.

One available choice is the set of output skims from the base scenario from which the analysis scenario is generated. In many cases, these will be the best choice for the seed skims; in some cases, however, it might be more beneficial to choose a different set of seed skims if the analysis scenario is more like a different previously run scenario than the base scenario. For example, if the analysis scenario is for a forecast year several years after the base scenario, another scenario that has been run for the same forecast year might be more efficient.

Future mobility settings

SERPM 8 has several new capabilities to represent future mobility, as described in the Future Mobility Support page. Examples of application of the initial developments of these capabilities with the SERPM 7 model are available here.

As the future mobility experiments are completed with SERPM 8, results and guidance for application will be made available.

Geographic sampling

Guidance and templates for utilizing the geographic sampling capabilities of CT-RAMP are under development.