Using Linked Non-Home-Based Trips in Virginia Hadi Sadrsadat, PhD, RSG Vince Bernardin, PhD, RSG May 15, 2017

Objectives

Additional Options for Agencies Main option for dealing with NHB trips has been just to develop an entirely new activity-based model According to a TMIP survey at the end of 2013: Near term, 3/4 of agencies will not have an activity-based model Long term, 1/3 – 2/3 of agencies will stay with trip-based models As TRB Special Report 288 states: “there is no single approach to travel forecasting or set of procedures that is ‘correct’ for all applications or all MPOs.”

The Challenge: The Problems with NHB trips Fundamental problems with NHB trips in trip-based models: When four-step models evolved in the 1950s & 1960s, NHB trips only accounted for ~15% of trips According to surveys, NHB trips now 30% of all trips By definition, NHB trips are disconnected from the households that make them and, thus, from the household’s HB trips This disconnect affects all dimensions of travel – location, mode, & time NHB trips are like noise in most models – pumped up to produce the right amount of overall traffic – but adding error to results

Spatial Problems with NHB Trips Physically impossible pattern for NHB trip patterns in four-step models: Travelers/vehicles appear and disappear, making trips between places they never went in the first place Unreasonable responses by NHB trips: New households can generate new NHB trips in remote locations they almost never visit on home-based trips

Other Problems with NHB Trips Mode and Time of NHB trips NHB trip modes are not conditional on HB trip mode choice in 4-step models: Shifting both HB and NHB trips to transit in transit improvement Decreasing HB & NHB walk trips in transit improvement In reality, more HB transit trips can lead to more NHB walk trips

Methodology

Understanding the Problem Four step models represent all trips through two component models: trip generation and trip distribution These correspond to the decision of whether/how frequently to make a trip and where to go (destination) These two models cannot properly define a NHB trip A NHB trip requires a third piece of information/decision: the origin, where to visit the destination from: 2 spatial (distribution) models are necessary to represent a NHB trip because neither its origin or destination are known

Background of New Methodology A Simple Solution: New methods explored by FHWA’s Travel Model Improvement Program (TMIP) Successfully implemented in Salt Lake City by RSG Very promising results with more reasonable responses to land use scenarios Simplest version of the method was used for this project

Linking NHB Trips to HB Trips Linking and running NHB distribution in series rather than in parallel with the HB distribution models HB & NHB distribution can assign both the origin and destination to NHB trips No changes to the HB model components

Generation by Mode Generation of NHB trips is done by mode, as a simple function of the HB attractions to that zone by each mode Mode choice is therefore unnecessary for NHB trips

Methodology Details NHB trips can be scaled to total NHB trips estimated in initial generation NHB trips are split to production and attraction equally The same friction factors were applied to NHB trips in distribution Explain that scaling is an option – done in Harrisonburg but not Fredericksburg

Model Implementations in Virginia

Harrisonburg Small city in Virginia 81,000 population NHTS, AirSage, and Streetlight data Typical 4-step model: 263 zones 5 trip purposes (HBW, HBS, HBO, HBU, NHB) 4 modes (DA, HOV, Transit, Non-motorizes) 4 time periods No destination choice Feedback loop

Fredericksburg Small city in Virginia, near DC 375,000 population NHTS, AirSage, and Streetlight data More transit ridership & slugging in I-95 express lanes corridor Typical 4-step model: 814 zones 5 trip purposes (HBW, HBS, HBO, HBU, NHB, IEW, IENW) 4 modes (DA, HOV2, HOV3+, Transit) 4 time periods No destination choice Feedback loop Mention, one of the few places in the US with slugging – special mode in mode choice

Charlottesville Small city in Virginia 150,000 population NHTS, AirSage, and Streetlight data UVA student population Advanced trip-based model: ~300 TAZ synthetic population disaggregate trip generation destination choice linked NHB trips accessibility variables perceived time CAV “plumbing” 4 time periods feedback loop In early stages of development, will incorporate other advances in trip-based modeling

Calibrated NHB Models Harrisonburg Simple regression models from household survey data NHB SOV = 0.251*HBW_SOV + 0.201*HBO_SOV + 0.120*HB_HOV (t=5.69) (t=8.11) (t=5.14) NHB HOV = 0.500*HBW_HOV + 0.239*HBO_HOV + 0.019*HB_SOV (t=3.44) (t=10.01) (t=1.05) HB SOV are most likely to generate NHB SOV, HB HOV more likely to generate NHB HOV, but both can generate either Note that this is a logical connection between the modes

Calibrated NHB Models Fredericksburg (under development) Different TOD models for HOV Also Transit NHB trips NHB TRN = 0.154*HB_TRN + 0.0008*HB_AUTO (18.25) (t=1.58) HB TRN are most likely to generate NHB TRN, but HB AUTO can generate a small number too Example: Person can get a ride to work, then take the bus somewhere else after work, etc.

Model Calibration NHB Trip Length Distribution Some inconsistency due to small sample NHTS data in Harrisonburg

Model Validation Old model – 1.98 New model – 2.96 NCHRP 716 – 3.0 Overall NHB Trip Rate per Household Old model – 1.98 New model – 2.96 NCHRP 716 – 3.0

Dynamic Validation / Sensitivity Testing New Residential Growth Important to test model response properties, not just base year replication Model response to new developments New hypothetical growth in town of Bridgewater SW of Harrisonburg No Development With Development Population 253 5000 Household 91 1800 Auto 170 3400

Original Model Response It would probably be better if we could get the NHB maps on the same scale as the HB

Enhanced Model Response It would probably be better if we could get the NHB maps on the same scale as the HB

Conclusions

Conclusions The enhanced Harrisonburg model with linked NHB trips performs better than the original model: More reasonable responses to hypothetical new residential growth Clearer, more intuitive and reasonable connections between NHB and HB modes Better able to replicate observed NHB trip rates, mode shares and OD patterns with less calibration

Contact Hadi Sadrsadat, PhD www.rsginc.com Hadi.Sadrsadat@rsginc.com Consultant Hadi.Sadrsadat@rsginc.com 240.283.0636