Nationally scalable and adaptive methods for traffic estimation with open data

A missing gap in the evidence base for transport planning, for many practitioners, is open data on road traffic: volumes, speeds, composition, timings, change over time. Knowledge of road traffic conditions, with confidence intervals to reflect uncertainty (for no dataset or model is perfect), is needed improve transport planning, especially in the context of active travel. Motor traffic volume and speed are associated with increased cycling and pedestrian injuries (Aldred 2019), hence impeding the uptake of active travel (Aldred and Goodman 2018) and the benefits it brings, such as improved mental and physical health and decarbonization (Department for Transport 2020). Therefore, it is not surprising that focus on active modes of travel has led to an increased demand for local evidence to inform interventions ranging from new cycleways to improved pavement quality.

This project will explore available datasets and tools to understand, prioritise and design active travel infrastructure, including cycleways, pavements, crossing points and traffic-calming features. Outputs will include new methods for extracting insight from a range of data sources on local provision (and gaps) for walking and cycling, 'source-agnostic' representations to capture policy relevant elements of infrastructure datasets and policy relevant findings on strengths and weaknesses in available datasets on active transport infrastructure in the UK. Given that it is a time of high demand for evidence-based sustainable transport policies, the project is expected to have international applications.

In terms of research background, it has been found that open data from OSM can provide a proxy for vehicle volumes (Chan and Cooper 2019). Another study on estimating traffic volumes have found that OD data analysis could produced desired results, especially after new 'jittering' methods to dis-aggregated OD datasets are used (Lovelace, Félix, and Carlino 2022).

The PhD will tackle the problem by developing new hybrid models, that combine machine learning, spatial network analysis, and origin-destination analysis to generate estimates of traffic volumes (and subsequently other variables such as speed and composition). Bayesian approaches will be explored, to provide confidence intervals in addition to central estimates of predicted traffic and impacts on road safety, building on recent work in the area (e.g. Gilardi et al. 2022). Finally, later steps will involve comparing the results with open access road crash data (Lovelace et al. 2019).

Expected outcomes:

Open estimates of road traffic at the link level
Central estimates of road traffic levels, measured in terms of PCU/day, at the link level joined to OSM road geometries
Confidence intervals representing the range within which observed values would be expected to fall 90% of the time (this could be followed-up by real world experiments to validate the model)
Open access code and models enabling updating of road traffic estimates nationwide as more data comes online (e.g. as road counter datasets are updated)
Academic papers
Exporatory analysis of road traffic data, e.g. looking at post COVID lockdown related impacts
On the methods underlying the work
Empirical data analysis of the results, e.g. comparing road traffic volumes with active travel infrastructure and/or behaviour
Outputs and activities directed by the DfT, e.g. tutorials for local authorities, reports on using the outputs from a policy perspective, and a workshop to build capacity with stakeholders on making use of traffic data for transport planning.
Policy changes resulting from the research