Identification of land-use characteristics using bicycle sharing data: A deep learning approach

Extensive research has shown that urban land-use characteristics, including resident, work, consumption, transit, etc., are significantly interrelated with travel behaviors and travel demands. Many research efforts have been made to evaluate the impact of land use planning or policies on travel behavior, however, few studies are able to quantitatively measure the land-use characteristics based on the data of travel behaviors or travel demand. In this paper, a new hybrid model that combines time series feature extraction and deep neural network is proposed to identify regional land use characteristics and quantify land use intensity using ridership data of bicycle sharing. This method consists of four main parts: (i) A set of land-use characteristic labels are evaluated based on planning and Geographic Information System (GIS) data. (ii) An ensemble clustering method is used to determine the segmentation points of ridership time series. (iii) The statistical characteristics of the segmented time series are extracted and used as input to the neural network. (iv) A deep neural network is established and trained based on the processed ridership features and land-use labels. In terms of data collection, ridership data of the bicycle-sharing parking spots and land-use planning data are obtained from bicycle-sharing system and planning department in San Francisco Bay Area, California U.S.A., respectively. The test results show that this approach has high accuracy for identifying land-use characteristics based on several standard evaluation measures and that the identification distribution can be well explained. The extension results further prove that the model can be applied to effectively analyze the main land-use characteristics of the region although the identification results may become unstable after 3–4 months.     

Extreme weather,public transport ridership and moderating effect of bus stop shelters

This paper examines how extreme weather conditions influence urban public transport ridership with a particular focus on the role of bus stop shelters. Using bus ridership data from the Salt Lake City metropolitan area, we find that extreme weather such as very high and low temperatures, and heavy rainfall reduces public transport ridership, while bus stop shelters have a modest effect on mitigating ridership losses resulting from these adverse weather conditions. The moderating effect of shelters is more pronounced on weekdays, and for bus stops with lower service frequency and fewer transfers. Our research also shows that the installation of bus shelters correlates with a variety of factors including service frequency, land use types, and local socioeconomic and demographic characteristics. Overall, our findings suggest that public transport amenities with weather-proof attributes have the potential to retain and attract more ridership on extreme weather days.     

Modeling bicycle crash costs using big data: A grid-cell-based Tobit model with random parameters

Bicyclists are among the most vulnerable road users in the urban transportation system. It is critical to investigate the contributing factors to bicycle-related crashes and to identify the hotspots for efficient allocation of treatment resources. A grid-cell-based modeling framework was used to incorporate heterogeneous data sources and to explore the overall safety patterns of bicyclists in Manhattan, New York City. A random parameters (RP) Tobit model was developed in the Bayesian framework to correlate transportation, land use, and sociodemographic data with bicycle crash costs. It is worth mentioning that a new algorithm was proposed to estimate bicyclist-involved crash exposure using large-scale bicycle ridership data from 2014 to 2016 obtained from Citi Bike, which is the largest bicycle sharing program in the United States. The proposed RP Tobit model could deal with left-censored crash cost data and was found to outperform the Tobit model by accounting for the unobserved heterogeneity across neighborhoods. Results indicated that bicycle ridership, bicycle rack density, subway ridership, taxi trips, bus stop density, population, and ratio of population under 14 were positively associated with bicycle crash cost, whereas residential ratio and median age had a negative impact on bicycle crash cost. The RP Tobit model was used to estimate the cell-specific potential for safety improvement (PSI) for hotspot ranking. The advantages of using the RP Tobit crash cost model to capture PSI are that injury severity is considered by being converted into unit costs, and varying effects of certain explanatory variables are accounted for by incorporating random parameters. The cell-based hotspot identification method can provide a complete risk map for bicyclists with high resolution. Most locations with high PSIs either had unprotected bicycle lanes or were close to the access points to protected bicycle routes.     

Ridership estimation of a new LRT system: Direct demand model approach

The successful introduction of LRT systems is inevitably related to the realistic estimation of their ridership; this is particularly true for cases of no prior experience in the use of such modes in the part of the traveling public. This paper presents a practical approach for developing a direct demand model, for the case of a planned LRT system in Cyprus connecting the three major cities of Nicosia, Larnaca and Limassol. The proposed approach is based on existing traffic demand data and limited roadside surveys. Results indicate that the introduction of the proposed LRT would attract a moderate number of 23,000 passengers daily and shift a small percentage of 3.5% of traffic to the system. It was also found that approximately 33% of these trips correspond to the urban section of the network, while about 62% of the estimated ridership will use the part of the system connecting Nicosia and Larnaca.     

Examining the influence of stop level infrastructure and built environment on bus ridership in Montreal

We studied transit ridership from the perspective of the transit provider, with the objective of quantifying the influence of transit system operational attributes, transportation system infrastructure attributes and built environment attributes on the disaggregate stop level boardings and alightings by time of day for the bus transit system in the Montreal region. A Composite Marginal Likelihood (CML) based ordered response probit (ORP) model, that simultaneously allows us to incorporate the influence of exogenous variables and potential correlations between boardings and alightings across multiple time periods of the day is employed. Our results indicate that headway affects ridership negatively, while the presence of public transportation around the stop has a positive and significant effect. Moreover, parks, commercial enterprises, and residential area, amongst others, have various effects across the day on boardings and alightings at bus stops. An elasticity analysis provides useful insights. Specifically, we observe that the most effective way to increase ridership is to increase public transport service and accessibility, whereas enhancements to land use have a smaller effect on ridership. The framework from our analysis provides transit agencies a mechanism to study the influence of transit accessibility, transit connectivity, transit schedule alterations (to increase/reduce headway), and land-use pattern changes on ridership.     

Exploring the impacts of land use by service coverage and station-level accessibility on rail transit ridership

In this study, we employ spatial regression analysis to empirically investigate the impacts of land use, rail service coverage, and rail station accessibility on rail transit ridership in the city of Seoul and the surrounding metropolitan region. Our analyses suggest that a rail transit service coverage boundary of 500 m provides the best fit for estimating rail transit ridership levels. With regard to land use, our results confirm that density is positively related to rail transit ridership within a 750 m radius of each station. In contrast, land use diversity is not associated with rail transit ridership. We also found that station-level accessibility is as important as land use for explaining rail transit ridership levels. Finally, we conclude that development density and station-level accessibility measures such as the number of station entrances or exits and the number of bus routes at the station are the most important and consistent factors for promoting rail transit ridership.     

Analysis on bike-share ridership for origin-destination pairs: Effects of public transit route characteristics and land-use patterns

Studies on bike-share programs have dramatically increased during the past decades. While numerous studies have examined various factors affecting bike-share demand at the station-level, few attempts have been made to understand bike-share ridership at the origin-destination (OD) level due to technical difficulties. The objective of this study is to examine whether existing public transit characteristics affect bike-share ridership at OD-level. We combined three datasets: (1) bike-share ridership data, (2) land-use and bike-transit infrastructure, and (3) bike-transit route characteristics between OD pairs of bike stations. Zero-inflated negative binomial (ZINB) regression models were used for the analysis. Our results showed that the travel distance between OD bike stations, land-use compositions, and the existence of bike-friendly infrastructures were significant factors determining bike-share ridership at the OD-level. In particular, a longer duration of public transit trips than bike-share, and more transit transfers, were associated with bike-share ridership. Further, this study showed that bike-share and public transit might compete with or promote each other, even within the city. The study's findings suggest that the relative efficiency of bike-share compared to public transit is highly associated with bike-share demand and help to increase the utility of bike-share system in response to several limitations of existing public transit networks.     

Modeling AADT on local functionally classified roads using land use,road density,and nearest nonlocal road data

The focus of this research is to model the influence of road, socioeconomic, and land-use characteristics on local road annual average daily traffic (AADT) and assess the model's predictability in non-covered location AADT estimation. Traditional ordinary least square (OLS) regression and geographically weighted regression (GWR) methods were explored to estimate AADT on local roads. Ten spatially distributed counties were considered for county-level analysis and modeling. The results indicate that road density, AADT at the nearest nonlocal road, and land use variables have a significant influence on local road AADT. The GWR model is found to be better at estimating the AADT than the OLS regression model. The developed county-level models were used for estimating AADT at non-covered locations in each county. The methodology, findings, and the AADT estimates at non-covered locations can be used to plan, design, build, and maintain the local roads in addition to meeting reporting requirements. The prediction error is found to be higher at urban areas and in counties with a smaller number of local road traffic count stations. Recommendations are made to account for influencing factors and enhance the local road count-based AADT sampling methodology.     

Spatially varying impacts of built environment factors on rail transit ridership at station level: A case study in Guangzhou,China

Understanding the relationship between the rail transit ridership and the built environment is crucial to promoting transit-oriented development and sustainable urban growth. Geographically weighted regression (GWR) models have previously been employed to reveal the spatial differences in such relationships at the station level. However, few studies characterized the built environment at a fine scale and associated them with rail transit usage. Moreover, none of the existing studies attempted to categorize the stations for policy-making considering varying impacts of the built environment. In this study, taking Guangzhou as an example, we integrated multi-source spatial big data, such as high spatial resolution remote sensing images, points of interest (POIs), social media and building footprint data to precisely quantify the characteristics of the built environment. This was combined with a GWR model to understand how the impacts of the fine-scale built environment factors on the rail transit ridership vary across the study region. The k-means clustering method was employed to identify distinct station groups based on the coefficients of the GWR model at the local stations. Policy zoning was proposed based on the results and differentiated planning guidance was suggested for different zones. These recommendations are expected to help increase rail transit usage, inform rail transit planning (to relieve the traffic burden on currently crowed lines), and re-allocate industrial and living facilities to reduce the commute for the residents. The policy and planning implications are crucial for the coordinated development of the rail transit system and land use.     

A GPS data-based analysis of built environment influences on bicyclist route preferences

This study examines the effects of built environment features, including factors of land use and road network, on bicyclists' route preferences using the data from the city of Seattle. The bicycle routes are identified using a GPS dataset collected from a smartphone application named “CycleTracks.” The route choice set is generated using the labeling route approach, and the cost functions of route alternatives are based on principal component analyses. Then, two mixed logit models, focusing on random parameters and alternative-specific coefficients, respectively, are estimated to examine bicyclists' route choice. The major findings of this study are as follows: (1) the bicycle route choice involves the joint consideration of convenience, safety, and leisure; (2) most bicyclists prefer to cycle on shorter, flat, and well-planned bicycle facilities with slow road traffic; (3) some bicyclists prefer routes surrounded by mixed land use; (4) some bicyclists favor routes which are planted with street trees or installed with street lights; and (5) some bicyclists prefer routes along with city features. This analysis provides valuable insights into how well-planned land use and road network can facilitate efficient, safe, and enjoyable bicycling.     

Investigating the associations between road network structure and non-motorist accidents

Road networks channel traffic flow and can impact the volume and proximity of walking and bicycling. Therefore, the structure of road networks—the pattern by which roads are connected—can affect the safety of non-motorized road users. To understand the impact of roads’ structural features on pedestrian and bicyclist safety, this study analyzes the associations between road network structure and non-motorist-involved crashes using data from 321 census tracts in Alameda County, California. Average geodesic distance, network betweenness centrality, and an overall clustering coefficient were calculated to quantify the structure of road networks. Three statistical models were developed using the geographically weighted regression (GWR) technique for the three structural factors, in addition to other zonal factors including traffic behavior, land use, transportation facility, and demographic features. The results indicate that longer average geodesic distance, higher network betweenness centrality, and a larger overall clustering coefficient were related to fewer non-motorist-involved accidents. Thus, results suggest that: (1) if a network is more highly centered on major roads, there will be fewer non-motorist-involved crashes; (2) a network with a greater average number of intersections on the shortest path connecting each pair of roads tends to experience fewer crashes involving pedestrians and bicyclists; and (3) the more clustered road networks are into several sub-core networks, the lower the non-motorist crash count. The three structural measurements can reflect the configuration of a network so that it can be used in other network analyses. More information about the types of road network structures that are conducive to non-motorist traffic safety can help to guide the design of new networks and the retrofitting of existing networks. The estimation results of GWR models explain the spatial heterogeneity of correlations between explanatory factors and non-motorist crashes, which can support regional agencies in establishing local safety policies.     

The influence of attitudes on university bicycle commuting: Considering bicycling experience levels

This study examined attitudes towards bicycling among bicycle users with different experience levels and how these attitudes influence the bicycle use. The study area is The Ohio State University’s main campus and we used the 2015 Campus Transportation Survey that asked questions about different commuting modes to the campus, bicycling experience levels, attitudes toward bicycle use, and demographic characteristics. For the empirical analysis, we grouped 20 attitudinal statements on bicycling into eight factors by using principal component analysis: perception of living in a bicycle-friendly community; perception of bicycling barriers; bicycling willingness upon facility availability; awareness of bicycling benefits; familiarity with local bicycling information; preference for bicycling; sensitivity to bicycle security; and perception of the availability of campus bicycle facilities. We ran t-test analyses to examine whether the attitudes toward bicycling vary by bicycling experience levels. Then, we employed binary logit analyses to estimate the effects of the attitudes differentiated by bicycling experience levels on being a bicyclist. The empirical analyses show that experienced bicycle users have more positive and favorable attitudes toward bicycling while less experienced bicycle users perceive greater bicycling barriers. We also found that the availability of bicycle facilities has a greater importance for less experienced bicycle users than for experienced bicycle users.     

Macro-level analysis of bicycle safety: Focusing on the characteristics of both crash location and residence

Over the last decade, bicycle ridership has been encouraged as a sustainable mode of transportation as it is economic and has less impact on the environment. Still, higher crash risk for bicyclists remains a deterrent for people to choose bicycling as their major mode of travel. As a first step in investigating bicycle safety, it is essential to identify not only the characteristics of the areas with the excessive number of bicycle crashes; but also those of the areas where crash-prone bicyclists reside. Therefore, this study aims to identify contributing factors for two subjects: (1) the number of bicycle crashes in the crash location's ZIP code and (2) the number of crash-involved bicyclists in their residence's ZIP. In order to achieve these objectives, a multivariate Bayesian Poisson lognormal CAR (conditional autoregressive) model was developed to identify the contributing factors for each subject. Regarding the model performance, the multivariate model outperformed its univariate counterpart in terms of DIC (deviance information criterion). Subsequently, hot zones for the two target zones were identified based on the modeling results. It is expected that practitioners are able to understand the contributing factors for bicycle crashes and identify hotspots from the results suggested in this study. In addition, they could implement safety countermeasures for the identified problematic locations to effectively reduce bicycle crashes.     

Influence of the built environment on E-scooter sharing ridership: A tale of five cities

Electric scooter (e-scooter) sharing systems (ESSs) have been widely adopted by many cities around the world and have attracted a growing number of users. Although some studies have explored the usage characteristics and effects of the built environment on ESS ridership using one city as an example, few studies have considered multiple cities to obtain generalizable and robust results. To fill this research gap, we collect the ESS trip data of five cities in the U.S., namely Austin, Minneapolis, Kansas City, Louisville, and Portland, and explore the effects of the built environment on ESS ridership after controlling for socioeconomic factors. The temporal distributions of e-scooter ridership of different cities are similar, having a single peak period on weekdays and weekends between 11:30 and 17:30. In terms of spatial distribution, the ESS ridership is higher in universities and urban centers compared to other areas. Multilevel negative binomial model results show that ESS trips are positively correlated with population density, employment density, intersection density, land use mixed entropy, and bus stop density in the census block group. E-scooter ridership is negatively correlated with the median age of the population in the census block group and distance to the city center. The findings in this article can help operators understand the factors that affect the ridership of shared e-scooters, determine the changes in ridership when the built environment changes, and identify high-ridership areas when ESS is implemented in new cities.     

Multi-city exploration of built environment and transit mode use: Comparison of Melbourne,Amsterdam and Boston

The built environment is an important determinant of travel demand and mode choice. Establishing the relationship between the built environment and transit use using direct models can help planners predict the impact of neighborhood-level changes, that are otherwise overlooked. However, limited research has compared the impacts of the built environment for different networks and for individual transit modes.This paper addresses this gap by developing built environment and transit use models for three multimodal networks, Amsterdam, Boston and Melbourne, using a consistent methodology. A sample of train, tram and bus sites with similar station-area built environments are selected and tested to establish if impacts differ by mode. It is the first study that develops neighborhood-level indicators for multiple locations using a consistent approach.This study compares results for ordinary least squares regression and two-stage least squares (2SLS) regression to examine the impact of transit supply endogeneity on results. Instrumented values are derived for bus and tram frequency in Melbourne and bus frequency in Boston. For other mode and city combinations, the 2SLS approach is less effective at removing endogeneity.Results confirm that different associations exist between the built environment and transit modes, after accounting for mode location bias, and that this is true in multiple networks. Local access and pedestrian connectivity are more important for bus use than other modes. Tram is related to commercial density. This finding is consistent for all samples. Land use mix and bicycle connectivity also tend to be associated with higher tram use. Train use is highest where opportunities exist to transfer with bus. Population density is commonly linked to ridership, but its significance varies by mode and network.More research is needed to understand the behavioral factors driving modal differences to help planners target interventions that result in optimal integration of land use with transit modes.     

Land use characteristics of subway catchment areas and their influence on subway ridership in Seoul

This study has dual research objectives: 1) to evaluate the land use characteristics of the pedestrian catchment areas (PCA) of subway stations in the Seoul metropolitan area (SMA) in terms of transit-oriented development (TOD) principles and 2) to investigate the influence of each PCA's land use characteristics on station-level ridership. The major findings can be summarized as follows. First, the built environments of subway PCAs in Seoul were found to be compatible with TOD principles in terms of density and diversity. They have declining density gradients of population and employment that extend outward from a station and have a high level of mixed-use land.Second, population and employment densities, land use mix diversity, and intermodal connectivity all have a positive impact on subway ridership, but differ in their spatial ranges. In particular, the influence spans of residential and commercial development patterns (population density and the proportion of commercial land) and mixed land use on ridership were limited to only the core and primary PCAs. Consequently, with respect to TOD in a compact city like Seoul, we recommend that a PCA be defined to encompass a radius of 600 m.     

Catchment if you can: The effect of station entrance and exit locations on accessibility

The success of passenger railway systems depends on their ridership and thus the population they serve. A mechanism to increase ridership is to expose the existing system to more people by reconfiguring the station itself, for instance by adding extra entrance and exit gates to shorten the walking distance from a trip's origin or its final destination. Gates are key nodes giving pedestrians access from street network to boarding/alighting facilities and vice versa. Stations and platforms are places not points, passengers may spend up to 6 min a trip walking between platforms and the end of the station nearest their origin or destination. This study systematically evaluates the accessibility of train stations and the effect of constructing an additional ‘far-side’ gate at stations with a single ‘near-side’ entrance. A three-step approach is defined to generate an isochrone as the catchment area for any transport node. Results indicate that stations with a single gate along their platforms (usually on one end of them) have the potential to increase the accessibility to jobs and population by around 10% on average. Due to the walking network and land use characteristics, some stations will benefit more significantly by retrofitting a new gate. Also, four linear regression models are developed to illustrate the effect of expanded accessibility on the number of entries and exits at each station for two peak periods. Then, stations are ranked based on their added ridership, which can help authorities to prioritize the development and allocating resources.     

Prioritizing low-stress and equitable bicycle networks using neighborhood-based accessibility measures

As cities across the United States strive to create comfortable bicycle networks for mainstream users, three topics have garnered attention: project prioritization criteria, accessibility to everyday destinations, and social equity. However, these topics have not often been integrated in research or practice. This paper introduces a method to assess the extent to which reductions in “Bicycle Level of Traffic Stress” (LTS) on segments of a citywide bicycle network increase accessibility to supermarkets, pharmacies, banks, and public libraries. Six accessibility performance measures are developed and evaluated for 278 neighborhoods in Baltimore, Maryland using a GIS-based approach. The demographic distributions of accessibility results are further analyzed, focusing on disadvantaged populations. Using a set of 106 proposed bicycle projects, the marginal accessibility gains and cumulative demographic impact across affected neighborhoods are assessed for each project. These results are ranked and crosslisted to identify a set of projects that balance accessibility gains with equity objectives. The prioritization results demonstrate some overlap with the short-term priorities embodied in City of Baltimore's 2015 Bike Master Plan, but they also highlight projects in other areas, specifically those that would serve neighborhoods most disadvantaged in terms of racial segregation, high poverty rates, and low rates of vehicle ownership.     

Residential dissonance and mode choice

Residential dissonance refers to the mismatch in land-use patterns between individuals’ preferred residential neighbourhood type and the type of neighbourhood in which they currently reside. Current knowledge regarding the impact of residential dissonance is limited to short-term travel behaviours in urban vs. suburban, and rural vs. urban areas. Although the prevailing view is that dissonants adjust their orientation and lifestyle around their surrounding land use over time, empirical evidence is lacking to support this proposition. This research identifies both short-term mode choice behaviour and medium-term mode shift behaviour of dissonants in transit oriented development (TODs) vs. non-TOD areas in Brisbane, Australia. Natural groupings of neighbourhood profiles (e.g. residential density, land use diversity, intersection density, cul-de-sac density, and public transport accessibility levels) of 3957 individuals were identified as living either in a TOD (510 individuals) or non-TOD (3447 individuals) areas in Brisbane using the TwoStep cluster analysis technique. Levels of dissonance were measured based on a factor analysis of 16 items representing the travel attitudes/preferences of individuals. Two multinomial logistic (MNL) regression models were estimated to understand mode choice behaviour of (1) TOD dissonants, and (2) non-TOD dissonants in 2009, controlling for socio-demographics and environmental characteristics. Two additional MNL regression models were estimated to investigate mode shift behaviour of (3) TOD dissonants, and (4) non-TOD dissonants between 2009 and 2011, also controlling for socio-demographic, changes in socio-demographic, and built environmental factors. The findings suggest that travel preference is relatively more influential in transport mode choice decisions compared with built environment features. Little behavioural evidence was found to support the adjustment of a dissonant orientation toward a particular land use feature and mode accessibility they represent (e.g. a modal shift to greater use of the car for non-TOD dissonants). TOD policies should focus on reducing the level of dissonance in TODs in order to enhance transit ridership.     

Modeling bike-sharing demand using a regression model with spatially varying coefficients

As an emerging mobility service, bike-sharing has become increasingly popular around the world. A critical question in planning and designing bike-sharing services is to know how different factors, such as land-use and built environment, affect bike-sharing demand. Most research investigated this problem from a holistic view using regression models, where assume the factor coefficients are spatially homogeneous. However, ignoring the local spatial effects of different factors is not tally with facts. Therefore, we develop a regression model with spatially varying coefficients to investigate how land use, social-demographic, and transportation infrastructure affect the bike-sharing demand at different stations to address this problem. Unlike existing geographically weighted models, we define station-specific regression and use a graph structure to encourage nearby stations to have similar coefficients. Using the bike-sharing data from the BIXI service in Montreal, we showcase the spatially varying patterns in the regression coefficients and highlight more sensitive areas to the marginal change of a specific factor. The proposed model also exhibits superior out-of-sample prediction power compared with traditional machine learning models and geostatistical models.