Institute for Transport Studies, University of Leeds

Active transport can tackle some of the most pressing issues of the 21^st century, including air pollution, the obesity epidemic, and transport-related social exclusion (Lucas 2004; Johansson et al. 2017; Saunders et al. 2013). Recognising these wide ranging benefits, the government has ambitious targets for active transport, as outlined in the Cycling and Walking Investment Strategy (CWIS): to double the number of stages cycled compared with the baseline year of 2013, and “reverse the decline in walking” (Department for Transport 2017). Notably, the government aims to double cycling whilst simultaneously reducing the number of people killed or seriously injured per billion km cycled year-on-year.

Data on cycling rates and future investment plans suggest that the cycling uptake target will be difficult to achieve (Road Safety GB 2019). Much research funded in support of the CWIS, such as the Propensity to Cycle Tool (Lovelace et al. 2017), has focussed on boosting walking and cycling levels, with relatively little attention paid to safety, despite the fact that cycling and walking casualties have not fallen since the CWIS baseline year of 2013 (see Figure 1). Safety is a major barrier to greater uptake of cycling and walking (Horton 2016; Jacobsen, Racioppi, and Rutter 2009), with fear of collisions and “too much traffic” being the top two reasons cited in DfT research (Nathan 2019).

The research outlined in the previous paragraph suggests that road safety measures can be an effective way of meeting both the active travel uptake and the road safety targets in the CWIS. This raises the question: which road safety measures work best to protect pedestrians and cyclists? Recent research has found that a range of traffic calming measures can reduce casualty rates, including ‘20 mph limits’ (implemented only via signalling) and ‘20 mph zones’ (which can include optional measures including speed cameras) (Maher 2018; ROSPA 2019), ‘filtered permeability’ interventions including and rising and fixed bollards, and a range of speed humps. A range of additional traffic calming measures is described in the DfT’s Local Transport Note (LTN) 01/07 (Department for Transport 2007). The focus of this study will be on three types of intervention, because these have received attention from the perspective of active travel (Mulvaney et al. 2015), and because data on their spatial (and in some cases date of implementation) is relatively abundant:

• 20 mph zones, which are well documented in a recent study by Maher (2018), and local case studies published for Bristol (Bornioli et al. 2018) and Greater London (Sarkar, Webster, and Kumari 2018).

  Data on the spatial extent, attributes and time of

  implementation of 20 mph zones will be collated from a range of sources, including via .gov.uk websites, newspaper articles and OpenStreetMap, building on research estimating the accuracy of crowd-sourced road data (Barrington-Leigh and Millard-Ball 2017).

• Physical traffic calming measures, including the location of speed cameras and speed hump data from local authorities OpenStreetMap.

• Filtered permeability, which will be measured on a sliding scale reporting connectivity for pedestrians and cyclists, relative to motorised modes.

Reproducible code that accesses, cleans and stores intervention data will be published in an R package, with the working title of trafficcalmr. This will lead to a step change in knowledge on how to access and process historic road calming intervention data, filling an important gap in current knowledge, and providing a robust evidence base on which future interventions can be prioritised and monitored.

Exposure estimates of pedestrians and cyclists to road danger has long been seen as a difficult component of road safety research. A report funded by the AA Foundation for Road Safety Research stated that “hard facts are lacking about the amount and pattern of activity which takes place each day in the road environment by pedestrians according to their age, gender and social background” (Ward et al. 1994). This is still largely true today, but data availability and processing capabilities have improved greatly, allowing estimates of walking and cycling levels for specific trip purposes to be estimated down to the route segment level. We will build on our previous research estimating exposure at area levels for road safety research (Lovelace, Roberts, and Kellar 2016) and extend it down to the road segment level using methods developed for the Propensity to Cycle Tool, but using higher resolution data, walking routing algorithms to estimate overall active transport levels, to create ‘heatmap’ estimates of exposure.

SaferActive will be a digital technology project using open data, with no direct impact on physical infrastructure, so to some extent road traffic calming regulations do not apply. We have good knowledge of guidance and the relevant regulations regarding road and street design and will use this to inform the scenarios of change represented in the application. Furthermore, users of the SaferActive web application should be aware of the relevant guidance. We will direct users towards guidance on integrated planning for urban transport contained in the Design Manual for Urban Roads (Lahart et al. 2013) and the recent book by John Parkin on designing for cycle traffic (Parkin 2018) to ensure interventions are done as part of joined-up transport plans. On specific road traffic measures, we will direct people towards the Local Transport Note 01/07 (Department for Transport 2007), the Design Manual for Roads and Bridges (DMRB) and guidance on 20 mph zones in the House of Commons Briefing paper CBP00468.

The project fully complies with ethical standards at the University of Leeds. The project will exclusively use open data, meaning that GDPR will not apply. There will be no personal data and the STATS19 data is already anonymised.

We are aware of the sensitive nature of some road safety datasets. The main datasets used by the project will be geographic aggregates. If for unforseen reasons we needed to work with sensitive data on the project, we would solicit advice from the Data Owner, the Data Protection Officer in University of Leeds’ Faculty of Environment and the funder before proceeding.

The structure of the project over its 18 month duration is shown in Figure 2. There are 3 main stages: (1) data access, cleaning and processing; (2) analysis and modelling; and (3) deployment of results. Key milestones will be the completion of the data processing stage (month 9),

release of a publicly available web application (month 12), and the workshops scheduled for months 10 and 17.

As a digital technology project delivered by an experienced team, the risks involved in funding this proposal are relatively small. Three main risks, and mitigating actions, are outlined in the table below.

Dr Robin Lovelace is an Academic Fellow in Transport and Big Data at ITS. He has expertise in data science, with a focus on transport and geographic data, and is author of widely used books on the subject, including Efficient R Programming and Geocomputation with R. He has demonstrable experience delivering high impact work, notably as Lead Developer of the Propensity to Cycle Tool (PCT), which is being used to design strategic cycle networks in local authorities across England and Wales

(Lovelace et al. 2017).

Dr Samantha Jamson is Professor of Transport Psychology specialising in aspects of workload and distraction with a focus on how processing of information from the road environment affects road safety. She has expertise in the design and evaluation of innovative road safety measures, ranging from low cost engineering interventions to in-vehicle workload managers. She also has experience in the human factors of interface design and the guidelines which should be used to ensure information presentation is clear and appropriate to the task in hand. This project will use her specialist road safety knowledge at the interface between engineering and psychology.

Dr Roger Beecham is an expert in Geographic Data Science and Visualization. With a PhD from the world-leading centre for Information Visualisation giCentre, he has published award-winning papers in geo-spatial data visualization at the prestigious IEEE VIS conference. Roger’s current research focusses on the application of modern data science for both data analysis and communication. Roger will work closely with Layik and Robin in contributing to the project’s methodology, proposing visualization interface designs and exploring and evaluating new approaches for uncertainty representation.

Dr Layik Hama is a technology expert passionate about road safety worldwide. After completing his PhD in Computer Science he led the development of two startup projects, one of which was replicating (www.freebase.com) which was acquired by Google and is now Google Knowledge Graph. Layik has been based at Leeds Institute for Data Analytics (LIDA) since May 2018, working with Robin on open data and research to support sustainable transport and Roger on a visualization framework funded by the Turing Institute. He will be the technical lead, using cutting edge technology to deploy the results in a scalable web application.

A recent conference paper by three of the team members shows we have a track record of collaboration on data science for road safety research.[1] We have also collaborated on writing a peer-reviewed R package called stats19 for increasing public access to reproducible road safety research (Lovelace et al. 2019).

SaferActive will innovate in a number of areas, pushing the boundaries of applied data driven road safety research, and advancing the knowledge-base needed to effectively prioritise road safety schemes for pedestrians and cyclists. A highly innovative aspect of the project will be its use of four datasets (on exposure, casualties, roads, and historic interventions) that have never previously been combined at the national level. This will add a huge amount of value to existing open datasets and provide a unique opportunity to learn from previously tried interventions.

The machine learning methods will allow not only better insight into the impacts of past interventions on vulnerable road users, but also projections of how effective future interventions could be. After testing the performance of training datasets against independent test data, we will provide the results nationwide. This will lead to innovation indirectly: by providing vital evidence on the likely impacts of different schemes, local authorities and other stakeholders will be able to taylor interventions to maximise both road safety and active transport impacts of new schemes. The quantification of ‘filtered permeability’ will be another innovative aspect of the project.

The most innovative aspect, however, will be the publication of the outputs in an interactive web application, ensuring wide uptake and maximum impact. We believe these innovations can lead to a step change in the policy, practice and public debate of road safety for vulnerable road users.

[1] See https://github.com/Robinlovelace/stats19-gisruk.