Model Improvement Investigations for Regional Rail (formerly CRT)�

with guidance from

Outline

• Wasatch Front TDM Realities
• Needs for Improvement
• Key Terms
• Main Conclusions
• Improvements Impact on Boardings
• Sensitivity Tests
• Recommended Improvements for WF TDM v9.2
• Observations

Wasatch Front TDM Realities

• Station-level accuracy is unrealistic, because transit is only calibrated and validated at the mode level.
• An improvement attempt in WF TDM v8.3.2 was unsuccessful, but it provided valuable lessons learned.
• With the upcoming Activity-Based Model (ABM) project, the mode choice model will be re-estimated, offering another opportunity to address issues.

In Utah, CRT serves travel patterns more like a long LRT than traditional CRT elsewhere.

–Joel Freedman, RSG

Needs for Improvement

1. Boardings for regional rail in model are in low in Utah County, especially Provo Station, and are high in Davis County
2. Drive-to-transit distance for regional rail in model is high for Utah County, especially Orem Station.
3. Distance traveled on regional rail in model is low. (from FrontRunner Forward study)

Green: On-Board Survey

Blue: WF v9.1 TDM

Note: Model run recalibrated so that total CRT boardings match OBS

Need #1:

Boardings in Utah County are Low and Davis County are High

​

​

​

Provo difference almost accounts for all Utah County. If we fix Provo, do we fix the model?

Green: On-Board Survey

Blue: WF v9.1 TDM

Need #2:

Utah County Drive- Access Distance is High

​

​

​

Orem should not be that high. Are people going to Orem, when they should be going to Provo?

Green: On-Board Survey

Blue: WF v9.1 TDM

Need #3:

Distance Traveled on Rail is Low

​

​

For transit trips, unlike auto, we don’t calibrate to distance. Could we do something about this?

Key Terms

In-Vehicle Time (IVT)

The portion of a trip spent inside a vehicle, either driving or riding transit.

Run Factor

Calibration parameter used in transit path-building and assignment that adjusts perceived in-vehicle time (IVT) by mode.

Driver Picks Orem

Driver Picks Provo

Which station will driver choose?

Max Time - Drive-to-Transit

Maximum allowable drive time from a traveler’s origin to a transit access point (like a park-and-ride or rail station) before the model assumes the person would not reasonably choose transit.

Provo

Station

15 minutes

35 minutes

Alternative Specific Constant (ASC)

Calibration parameter in discrete choice models that captures the average preference for an alternative not explained by observed attributes.

• Adjusts model predictions to better match observed mode shares
• Accounts for unmeasured factors (comfort, image, reliability, policies, etc.)
• Transit ASC may capture elements like vehicle cleanliness, prestige, or perceived safety.

HBO Off-Peak Trip Length Frequency by 5-Mile Bins

Distance on CRT (miles)

Modeled

Observed

Main Conclusions

​

• Adjusted RUNFACTOR to 2.5 for drive-to-transit and drive access MaxTime align median drive-access distance for Utah County stations.
• Purpose/period-specific Alternative Specific Constants (ASC) by distance bins significantly improve distance-on-CRT distributions (from -11% to -56% error to -5% to -18%).
• Magnitude of adjustments needed for Provo HBW boardings to match observed is 60-75 minutes of in-vehicle time for southern Utah County. However, model sensitivity is unsatisfactory at this level of increase.

Boardings Impacts of Improvements

P

P

A

A

O

D

D

O

PA (Production / Attraction)

OD (Origin / Destination)

boarding

boarding

alighting

alighting

boarding

boarding

alighting

alighting

PA boardings will show two boardings at station of production (home for Home-Based Work)

Real World Trips

Modeled World Trips

0% Variance GOAL

2019 Transit On-Board Survey

Note: Model run recalibrated so that total CRT boardings match OBS

0% Variance GOAL

WF-TDM

v9.1 Model Transit Recalibrated

​

Note: Model run recalibrated so that total CRT boardings match OBS

0% Variance GOAL

Run Factor Increase

​

Increase from 1.5 to 2.5

Note: Model run recalibrated so that total CRT boardings match OBS

0% Variance GOAL

Alternative Specific Constants by Distance

​

Individual constants by 5-mile bins

Note: Model run recalibrated so that total CRT boardings match OBS

0% Variance GOAL

Utah County Max Drive-to-Transit Time of 15 minutes

Note: Model run recalibrated so that total CRT boardings match OBS

0% Variance GOAL

Utah County Transit In-Vehicle Time -15-minute adjustment

Note: Model run recalibrated so that total CRT boardings match OBS

0% Variance GOAL

Southern Utah County Transit In-Vehicle Time -30-minute Adjustment

Note: Model run recalibrated so that total CRT boardings match OBS

0% Variance GOAL

Southern Utah County Transit In-Vehicle Time -45 minute adjustment

Note: Model run recalibrated so that total CRT boardings match OBS

0% Variance GOAL

Southern Utah County Transit In-Vehicle Time -60 minute adjustment

Sensitivity Tests

• 2019 Base
• 2019 Increase Frequency 15/30 minutes
• 2019 Extend regional rail south to Payson
• 2019 with 2050 Households and Jobs
• 2050 RTP Transit and Highway Projects

Sensitivity Tests ran for Models Modifications (additive)

• 1-AscDist: Recalibrate with ASCs by 5-mile bins
• 2-Max Time 15: CRT drive-access max reduced to 15 min (was 35; OBS 90th %ile)
• 3-UTCo -15 adj: −15 min IVT-equivalent for all Utah County productions
• 4-UTCoSouth -60 adj: −60 min IVT-equivalent for Provo & south Utah County productions

Sensitivity Tests

Provo Station PA Boardings

Increased Frequency Sensitivity Test

comparing to

base year

network and SE

Note: Modifications are additive, meaning each one includes previous modifications

What model modification did the best in this test?

Extend Rail to Payson Sensitivity Test

comparing to

base year

network and SE

Note: Modifications are additive, meaning each one includes previous modifications

What model modification did the best in this test?

2019 with 2050 HH and Jobs Sensitivity Test

comparing to

base year

network and SE

Note: Modifications are additive, meaning each one includes previous modifications

What model modification did the best in this test?

2050 RTP Sensitivity Test

comparing to

base year

network and SE

Note: Modifications are additive, meaning each one includes previous modifications

What model modification did the best in this test?

Recommended Improvements for WF TDM v9.2

1. Adjust Drive-Access Utility for CRT
• Set RUNFACTOR = 2.5 for drive-access links to regional rail.
• Better aligns modeled behavior with observed travel patterns, particularly at end-of-the-line stations (Provo and Ogden).
2. Refine Regional Rail Distance-on-Transit Modeling
• Apply Alternative Specific Constants (ASCs) by distance bins.
• Improves fit to observed distributions of trip distances on CRT.
3. Document Lessons Learned for v9.2 Calibration and Future Model Application
• Summarize non-implemented adjustments needed to bring modeled boardings in line with observed values at the county level and for Provo Station. Inform on magnitude of issue.
• Provide useful insights for those using the model in transit forecasting or future local studies.
• Support planning for v9.2 calibration using HTS 2023 and Transit Onboard Survey 2023 data.

Observations #1

• Model performs consistently across stations: Provo drive-to-CRT access is aligned with other stations in Utah and Southern Salt Lake Counties, based on household-to-boarding ratios within 3-mile buffers
• Observed data at Provo is an outlier: Walk-to-CRT boardings closely match observed values, but drive-to-CRT boardings reach only 10% of observed; tuning RUNFACTOR and distance-based ASC improves performance to 25%
• Provo boardings concurrence: Automated Passenger Count (APC) data matches on-board survey boardings system-wide, including Provo.
• CRT between SLC and Provo has a more balanced peak directional split than CRT between SLC and Ogden is more peak direction heavy (see next slide for figure)

Observations #2

• ASC Distance calibration had minimal impact on 2050 forecast—actually increases drive-to-CRT boardings by 7%.
• UrbanVal (development density) doesn't affect boarding pattern, improve Provo, or significantly reduce future projections.
• Removal of Park-N-Ride lots for Salt Lake Central and North Temple and recalibration did not improve Provo.
• Provo/South-Utah-County-to-SLC auto trips are being over predicted slightly by model not under predicted.
• Provo and Weber County are most impacted by In-Vehicle Time portion of utility equation relative to other components. (see next slide for figure)