Temporal Variations in Daily Activity Networks Using Smartcard Data

This study explores temporal variations in activity networks for four million passengers, differentiated as workers and non-workers, using public transport based on a large-scale smart card dataset generated over 105 days in Beijing. We aim to capture their day-to-day transition and cumulative temporal expansion in activity network using transit over days, weeks, and months. Particularly, workers and non-workers are automatically identified based on their different daily routines, whose activity networks are characterized by six features concerning space coverage, distance coverage, and frequency coverage in two ways, namely, on a per-day transition and with an accumulation of days. The transition features of the networks are statistically analyzed and compared by time, while how the expansion features evolve with time are modeled. Results show that, on weekdays, workers are more likely to travel longer (have larger distance coverage), but cover less area (have smaller space coverage) than non- workers. While opposite patterns occur on weekends. Traveling in the ‘North-South’ direction is weakly correlated with traveling in the ‘East-West’ direction. Workers on weekdays, as well as non-workers on weekends, make longer ‘North-South’ trips. Manhattan distance, trip count, and perimeter present a ∩ shape in their probability density functions, while the remaining features decline dramatically, with probability density functions fit by the exponential distribution. The distance coverage expands faster than that of space coverage. Most passengers increase coverage of space and distance when time expands (obviously no one decreases coverage over time, but some don’t change). The research enables findings on temporal load-balancing, long-term cumulative expansion in travel demands of workers and non-workers, re-balancing the distribution of existing workplace and residential location opportunities, and constructing transit-oriented developments with mixed functions over time.

1_5disAll1.png

A Review of Game Theory Models of Lane Changing

Recent working paper:

Driver lane-changing behaviours have a significant impact on the safety and the capacity of the vehicle-based traffic system, therefore, modeling lane-changing maneuvers has become an essential component of driving behaviour analysis. Among microscopic LC models, game theory based lane-changing models highlight the interaction or competition of drivers, which reveal a more realistic image of driving behaviours compared to other classic models. However, the potential of game theory to describe the human driver’s lane-changing strategies is currently under-estimated. This paper aims to review the recent development of game theoretic models that are classified according to their different methodologies and features. They are designed for both human-driven (User-optimal) and autonomous vehicles (User and system-optimal), and we hope they can find applications in future AV industries.

classGT

Trains, trams, and terraces: population growth and network expansion in Sydney: 1861-1931

Recent working paper:

This paper examines the changes that occurred in the tram and train networks and density of population in Sydney between the early 1860s and 1930s when both trains and trams were developing. A set of statistical analysis has been conducted using panel data representing 593 districts of Greater Sydney at suburb (neighborhood) level over each decade from 1861 to 1931. We find that trams and population density are positively associated in a positive feedback process, tram deployment leads population growth and population growth leads tram deployment, both satisfying a Granger causality test.

1921

Commute Mode Share and Access to Jobs across US Metropolitan Areas

Recently published:

How much of the variation in transit mode share is attributable to accessibility is not well understood, despite its significant policy implications. It is hypothesized that better transit accessibility leads to higher transit mode share. This paper explains block-group level transit mode share using transit accessibility in a logistic model for 48 major US metropolitan areas. Transit accessibility alone explains much of the variation in transit mode share for all 48 regions despite their geographical differences (adjusted R2 0.61, potential accessibility); models for individual cities have stable and interpretable parameters for transit accessibility. The models better explain mode share in cities with higher person weighted transit accessibility and larger populations; an adjusted R2 of 0.76 is achieved for New York City with transit accessibility as the only explanatory variable. Additional automobile accessibility and income variables modestly improve model fit. Time-decay functions fitted to accessibility measures better explain mode choice than the isochrone accessibility, and suggest the catchment area affecting transit mode choice to be within 35 minutes. This work contributes to the understanding of transit mode share by solidifying its link with accessibility, which is determined by the structure of the transport network and land development.

Boston30

Shortest paths, travel costs, and traffic

Recent working paper:

This study focuses on path flow for road network, as the sum of individual route choices from individual travelers, associated with specific path type for each cost fac- tor of auto travel that finds the optimal route with the minimum cumulative cost from the perspective of the corresponding cost analyst interest. The considered cost factors include time, safety, emission, and monetary costs, as well as their composite, internal and full cost of travel. We further explore the extent to which each cost factor explains the observed link traffic flows given an estimated home-to-work demand pattern. The results of the Minneapolis – St. Paul metropolitan area indicate that flows from multiple path types, associated with internal cost components, additionally to the factor of distance, provides the best fit.

 

dif_full_2

Moving Array Traffic Probes

Recent working paper:

  • Davis, Blake, Ji, Ang,  Liu, Bichen, and Levinson, D. (2020) Moving Array Traffic Probes. To be presented at the Transportation Research Board Annual Meeting, January 2020.

This paper explores the potential of moving array `probes’ to collect traffic data. This application simulates the prospect of mining environmental data on traffic conditions to present a cheap and potentially widespread source of traffic conditions. Based on three different simulations, we measure the magnitude and trends of probe error (comparing the probe’s `subjective’ or time-weighted perception with an `objective’ observer) in density, speed, and flow in order to validate the proposed model and compare the results with loop detectors. From these simulations, several conclusions were reached. A single probe’s error follows a double hump trend due to an interplay between the factors of traffic heterogeneity and shockwaves. Reduced visibility of the single probe does not proportionately increase the error. Multiple probes do not tend to increase accuracy significantly, which suggests that the data will be still useful even if probes are sparsely distributed. Finally, probes can measure the conditions of oncoming traffic more accurately than concurrent traffic. Further research is expected to consider more complex road networks and develop methods to improve the accuracy of moving array samples.

concept1

The Transit Travel Time Machine: Comparing Three Different Tools for Travel Time Estimation

Recent working paper:

Estimating the OD travel time matrix is an essential step in transport demand forecasting, accessibility analysis, spatial interaction modeling, and reliability analysis. Many tools have been developed to estimate the travel time between origins and destinations by different transport modes. This paper systematically reviews three tools: ArcGIS Network Analyst, OpenTripPlanner, and Google Maps Application Programming Interface, that are able to estimate an OD travel time matrix, and compares the results for 30 stations of Sydney Trains network with a same set of parameters. Bus, train (rail), and transit modes are tested. Results indicate that OpenTripPlanner produces higher transit travel times than Google Maps API (2.8 minutes) while ArcGIS Network Analyst estimates a travel time 6 minutes higher on average than Google Maps API for transit modes.

Fig4

What’s Access Worth? A Hedonic Pricing Approach to Valuing Cities

Recent working paper:

Access, or the ease of reaching destinations, influences property values. Most studies investigating the relationship between the two either confuse access for proximity, or measure accessibilities in ways that make interpreting elasticities difficult. We specify, and estimate, a hedonic pricing model for house sales in the Sydney region where access by auto and transit are measured as the cumulative number of jobs (opportunities) reachable within a determined time threshold, here 45 minutes. The model was set up with structural, and neighborhood attributes in addition to the access measures, and resulted in a statistically significant fit. The results indicate a greater influence of transit access on house prices compared with auto access. Although a synergy would play out in reality, a unit increase in jobs reachable within 45 minutes by transit was estimated to increase sale price by more than six times than auto. We also demonstrated the impact of improving access on the value of postal code areas. Although the results are specific to the Sydney region, the method can be adapted to any region to estimate elasticities of sale price with respect to access.

salePriceDistribution

How Transit Scaling Shapes Cities

Recently published:

Transit accessibility to jobs (the ease of reaching work opportunities with public transport) affects both residential location and commute mode choice, resulting in gradations of residential land use intensity and transit (public transport) patronage. We propose a scaling model explaining much of the variation in transit use (transit commuters per km2) and residential land use intensity with transit accessibility. We find locations with high transit accessibility consistently have more riders and higher residential density; transit systems that provide greater accessibility and with a larger base for patronage have proportionally more ridership increase per unit of accessibility. All 48 metropolitan statistical areas (MSAs) in our sample have a scaling factor less than 1, so a 1% increase in access to jobs produces less than 1% increase in transit riders; the largest cities have higher scaling factors than smaller cities, indicating returns to scale. The models, derived from a new database of transit accessibility measured for every minute of the peak period over 11 million US census-blocks, and estimated for 48 major cities (MSAs) across the United States, find that jobs within 45 minutes most affect transit rider density. The findings support that transit investment should focus on mature, well-developed regions.

Scaling Coefficients for Transit Commuter and Working Population Density (dot sizes corresponds to residential density); Between-cities scaling (pooling 48 cities) shown with the solid red dot; All cities above the red diagonal line
Scaling Coefficients for Transit Commuter and Working Population Density (dot sizes corresponds to residential density); Between-cities scaling (pooling 48 cities) shown with the solid red dot; All cities above the red diagonal line