To a machine, Hong Kong doesn’t look so different from Marrakech, Guadalajara or Port-au-Prince – at least, according to new urban design neural networks developed by a team of scientists led from the University of Melbourne’s Transport, Health & Urban Design (THUD) Research Hub .
Aiming to find links between city layouts and road carnage, researchers from THUD, with collaboration from Melbourne’s Earth Sciences Department, Barcelona Institute of Global Health and the Mailman School of Public Health at Columbia University, used neural nets and graph-based approaches to analyse the road, rail and waterway networks of 1,692 cities, covering a third of the world’s population.
Creating 1,000 images per city, the study worked with a dataset of around 1.7 million images to learn and model how layouts interact with deaths and injuries on the roads. The authors say this approach produces a degree of fidelity for understanding city types not previously possible.
Researchers found cities can be well categorised into just nine types; and that, adjusted for economic activity, the worst city types for road injuries show around twice the road carnage impact of the best performing city types.
Hong Kong emerges as a “chequerboard”, a type common in Latin America and northern Africa and typical, the researchers say, of cities developed in line with the “Laws of the Indies” (the style of Spanish colonies’ 16th century urban development, built on the ruins of conquered cities in line with a centralised urban tradition enforced by a distant king).
Chequerboards are also one the poorest performing types of city, in terms of transport safety: featuring high proportion of land use for roads and low land use for rail, they show a mean transport injury burden 1.93 times higher than the best performing city types.
Going beyond the car
One problem identified by the researchers is that many chequerboards, (and other poorly performing city types), are in low-income, fast-developing regions. These countries are often targeted by multilateral development bank infrastructure projects promoting “road safety”: yet such projects often fail to look at transportation as a whole when addressing death or injuries on roads.
“Seldom do these projects advocate investment in safer public or active transport alternatives beyond the car,” the researchers write in The Lancet Planetary Health.
But University of Melbourne’s Dr Jason Thompson, who led the imaging study, says even richer cities like Hong Kong need to make a shift from “road safety” to “transport safety”.
“We argue that transport safety is key, not road safety. For too long, everyone has been focussed on how to make the car safer. This is like trying to make low-tar cigarettes,” he says.
Thompson recommends Todd Litman’s Vision Zero, Meet VMT Reductions article as a good summary of the “old and new” traffic safety paradigms. So, for example, old solutions are roadway and vehicle design improvements; new solutions are road, parking, fuel and insurance pricing reforms.
“There is a growing realisation that we can’t continue to just build roads and solve the safety problem through the ‘old’ road safety paradigm,” says Thompson.
One limitation of the research, regarding Hong Kong, is that public transport waterways (ferries and water taxis) and bus networks are not included in the machine learning inputs. “We had to strip the maps back to core components that were consistently depicted across the world,” says Thompson. “Unfortunately this was not the case for ferries and buses.”
Adding these in might change classifications to some extent, he says, “but not hugely, as it becomes about the use of the city infrastructure rather than the infrastructure per se. Bicycles and pedestrians paths could also be thought of the same way – we hope to address these questions in the future as the maps become more consistent.”
Jason Thompson, Mark Stevenson, Jasper S Wijnands, Kerry A Nice, Gideon DPA Aschwanden, Jeremy Silver, Mark Nieuwenhuijsen, Peter Rayner, Robyn Schofield, Rohit Hariharan, Christopher N Morrison, A global analysis of urban design types and road transport injury: an image processing study, The Lancet Planetary Health, Volume 4, Issue 1, 2020, Pages e32-e42, ISSN 2542-5196, https://doi.org/10.1016/S2542-5196(19)30263-3.