Agenda
- Agency update: PSRC
- Draft Land Use Activities and Application Analysis Guide (CS)
Notes
Agency Update: PSRC (Stefan Coe and Bryce Nichols)
Background and History
PSRC began implementing ActivitySim around 2018, initially borrowing the MTC example model and adapting their land use and skim data to match its inputs. This early phase served as a learning opportunity with relatively low risk, as ActivitySim was not yet their production model. Over time, PSRC added features to reach parity with their legacy DaySim model (Soundcast), including a two-zone (TAZ/MAZ) implementation, multiprocessing, and Sharrow for runtime improvement.
RSG Support Contract (2023–2025) – Key Accomplishments
- Value of Time (VOT) categories: Added low/medium/high VOT segmentation for tour and trip mode choice, with corresponding skims and perception factors used in auto assignment.
- Disaggregate accessibilities: Tour mode choice logsums—segmented by income and auto ownership—are now used in auto ownership, work-from-home, CDAP, and non-mandatory tour frequency models. Implementation required careful handling of auto-sufficiency matching prior to the auto ownership model run.
- Model estimation: Multiple models re-estimated using PSRC's 2023 Household Travel Survey, with contributions from RSG staff and PSRC staff.
Current Development
- Parcel-level detail: PSRC generates MAZs from parcel inputs and converts model outputs back to parcel level using size terms and transit distance decay functions, preserving analytical capabilities for geographies like regional growth centers (e.g., Bellevue, now connected to light rail).
- Park-and-ride lot choice model: Moved from OpenPaths (~20 min/iteration) to an in-ActivitySim implementation (~1–2 min). Capacitated model is functioning as expected; estimation and calibration are still needed.
- Vehicle type choice: An intern (Mohamed) has initiated this work; seen as important for EV fleet forecasting.
- Input checker: Recently added; expected to be valuable when updating the base year to 2030.
Next Steps
- Complete calibration and finalize new models
- Run ActivitySim in shadow mode alongside DaySim for the current RTP
- Update base year to 2030 and apply auto-calibration tools
Q&A Highlights
- Joel asked about geocoding park-and-ride lot locations from survey data; PSRC is inferring lot locations from linked trip records and plans to present their methodology once estimation files are built.
- Bo raised a Python 3.12 compatibility issue affecting joint installation of ActivitySim and PopulationSim, and urged expediting the park-and-ride pull request (currently in review per David Hensel) to avoid further divergence.
- Joe C noted PSRC's model occupies an interesting middle ground between the SANDAG and MTC examples.
Draft Land Use Activities and Application Analysis Guide (Joe Flood)
Joe F shared a first draft of the land use scenario analysis guide. The draft uses a Transit-Oriented Development (TOD) scenario at Grossmont Center (adjacent to the Grossmont Trolley Station on the San Diego Green and Orange lines) as its worked example—adding 2,000 households and 1,000 retail jobs across two currently MAZs. The guide covers scenario setup (three files to modify), use of PopulationSim in repop mode, and model run procedures.
Initial Feedback
- Joe Castiglione noted the right elements are covered but flagged a potential need for greater granularity—particularly to support a junior modeler working through the scenario for the first time with the SANDAG example.
- A section on results interpretation (e.g., appropriate metrics, VMT per capita) is identified as a gap to address in a future revision, building on discussion from the PSRC presentation.
- Bhargava Sana's initial read was that all elements are present; more detailed review is forthcoming.
Review Process
The group discussed feedback mechanisms. Joe Flood proposed using a GitHub pull request; Bo Wen (TransLink) supported this approach, noting it is feasible for markdown and used for code review at TransLink (requires GitHub accounts). Joe Castiglione offered a shared Google Doc as an alternative. The group agreed to proceed with the GitHub pull request approach.
Agenda
Notes
Agency Update: PSRC (Stefan Coe and Bryce Nichols)
Background and History
PSRC began implementing ActivitySim around 2018, initially borrowing the MTC example model and adapting their land use and skim data to match its inputs. This early phase served as a learning opportunity with relatively low risk, as ActivitySim was not yet their production model. Over time, PSRC added features to reach parity with their legacy DaySim model (Soundcast), including a two-zone (TAZ/MAZ) implementation, multiprocessing, and Sharrow for runtime improvement.
RSG Support Contract (2023–2025) – Key Accomplishments
Current Development
Next Steps
Q&A Highlights
Draft Land Use Activities and Application Analysis Guide (Joe Flood)
Joe F shared a first draft of the land use scenario analysis guide. The draft uses a Transit-Oriented Development (TOD) scenario at Grossmont Center (adjacent to the Grossmont Trolley Station on the San Diego Green and Orange lines) as its worked example—adding 2,000 households and 1,000 retail jobs across two currently MAZs. The guide covers scenario setup (three files to modify), use of PopulationSim in repop mode, and model run procedures.
Initial Feedback
Review Process
The group discussed feedback mechanisms. Joe Flood proposed using a GitHub pull request; Bo Wen (TransLink) supported this approach, noting it is feasible for markdown and used for code review at TransLink (requires GitHub accounts). Joe Castiglione offered a shared Google Doc as an alternative. The group agreed to proceed with the GitHub pull request approach.