Through a variety of transport modes, e.g.,

ride-sourcing public transit private car

Transport modal disparities

• Carbon intensity

• Spatiotemporal distributions of travel time and trips

RQ1 What are the potentials and limitations of using emerging data sources for modelling mobility?

RQ2 How can new data sources be properly modelled for characterising transport modal disparities?

RQ	#	Scope	Paper title
1	I	Population heterogeneity	From individual to collective behaviours: exploring population heterogeneity of human mobility based on social media data
	II	Travel demand	Feasibility of estimating travel demand using geolocations of social media data
	III		A mobility model for synthetic travel demand from sparse individual traces
2	IV	Travel time	Disparities in travel times between car and transit: spatiotemporal patterns in cities
	V	Modal competition	Ride-sourcing compared to its public-transit alternative using big trip data

• Why Twitter?

  • Easy access, low cost, large spatial and population coverage.

• Limitations of geotagged tweets

  • Biased population: young, highly-educated, urban residents.

  • Sparse sampling of the actual mobility.

  • Behaviour bias of reporting geolocations.

• Twitter users DO NOT geotweet every day.

• Twitter users DO NOT geotweet every location visited.

Uncommon places and leisure activities more than regularly visited places, e.g., home and workplace.

The reliability of estimated commuting trips using geotagged tweets is low.

Commuting trip distance distributions

Four types of travellers

• Local vs. Global traveller visits

  • Local: nearby locations.

  • Global: more distant locations.

• Returner vs. Explorer explores around

  • Returner: one centralised location.

  • Explorer: decentralised locations that are distant from each other.

A density-based approach is proposed to increase sample size for estimating travel demand.

Trip-based approach in the literatureLee et al., 2019

Our density-based approach

increases the accuracy of the estimation by 10%–60%.

• An individual-based mobility model is proposed to fill the gaps in sparse mobility data.

• The model is designed to correct behaviour bias and sparsity issue.

Input- sparse mobility traces.

Output- synthesised mobility converted to daily trips.

• The model-synthesised results have good performance.

• An application: characterising trip distance distributions (domestic) of global regions’ residents:

• ⭕️ ❗️ Easy access, low cost, but with biased population, behaviour bias, and sparsity issue.

• 👤 At the individual level, fundamental patterns are preserved.

• 👥 At the population level:

  • a reasonably good source for the overall travel demand estimation but not commuting demand.

  • careful consideration on spatial scale, sampling method, and sample size.

• 🔧 Innovative approaches for correcting the biases and increasing available data.

Data fusion framework for travel time calculation:

Distribution of geotagged tweets represents the dynamic attractiveness of locations in cities.

Travel time ratio (R) by hour of day (Sydney)

Travel time ratio (R) over 24 hours

• Travel time by PT is around twice as high as by car.

• PT can compete with car use during peak rush hours in Stockholm and Amsterdam.

	Does ride-sourcing complement, or compete with, public transit?

• How large is the share of ride-sourcing trips that can be substituted by taking public transit, if you are willing to walk up to 800 m to access and leave the transit station during daytime?

• Ride-sourcing trips: transit-competing vs. non-transit-competing.

  • What trip attributes and built environment are linked to the competition?

  • What are the implications for policymaking?

• The transit-competing trips account for 48.2% of the overall 4.3 million ride-sourcing trips.

Hot spots of ride-sourcing trips

Raw data
trip ID
pick-up/drop-off locations
pick-up/drop-off times
cost

	Enriched variables
Trip attributes	🚌 public transit information, e.g., travel time
	☀️ weather condition
	🕸 demand-based communities (by trip-based network)
Built environment	🚏 transit-stop density
	🏨 functional regions (by points of interest)

• A glass-box model enhanced by machine learning techniques: additive impact of variables and variable interactions.

Land-use clusters of the study area

• Low density/diversity of land use -> a lower probability of competition.

Land-use clusters of the study area

• Multiple transfers + middle density/diversity of land use -> a higher probability of competition.

• 🚌 Expand PT networks guided by the transit-competing ride-sourcing trips featuring

  • short travel time by ride-sourcing (< 15 min)

  • large travel time ratio between the two modes;

• 💰 Incentivise the ride-sourcing trips that fill the gaps in the PT services which

  • take a long time

  • require lengthy walking

  • require multiple transfers connecting to suburban areas.

• 📦 Importance of data fusion approaches, especially given more and more open but incomplete data.

• Geotagged tweets is a good source for time-varying attractiveness of urban locations.

• 🚌 🚗 Public transit is virtually always slower than car and ride-sourcing.

• 📍 For making public transit more competitive, spatiotemporal details add nuanced insights to identify gaps and opportunities.

• Provide validation to identify the limitations of geotagged tweets: behavior bias, biased population, and sparsity issue.

• Reveal the potentials of geotagged tweets at both individual and population level.

• Reveal the spatio-temporal disparities between car/ride-sourcing and public transit about travel time and modal competition.

• Propose a density approach and an individual-based mobility model for travel demand estimation, addressing sparsity issue and behaviour bias of geotagged tweets.

• Create two reproducible data fusion frameworks integrating multiple data sources from transport systems for characterising modal disparities.

• Extending the use of social media data for mobility modelling.

• 🌐 Generating global synthetic mobility data for improving travel demand projections.

• 🚙 & 🚊 🚲 🚕 🚌 🚋 Combining multi-modal trip data to explore occupancy, shareability, and electrification of new mobility services for reducing transport carbon emissions.

• 🕸 Introducing the perspective of networks.