GHD partners on prize-winning mobility research
New approach to dynamic urban classification using mobility data
At a glance
GHD supported prize-winning research using mobility data to improve how urban areas are classified over time.
Prize-winning project uses mobility and POI data to map how urban areas change across daily and weekly patterns.
GHD’s Data and Insights team has supported a prize-winning masters project which develops a new way to characterise areas of a city based on the daily ebb and flow of visitors. Led by Chung-En Tsern from University College London and recognised in the Geographic Data Service masters competition, the project combined mobility patterns with points of interest (POIs) to sharpen urban classification by capturing how places change across the day and the week.
Data and Insights’ Cristobal Montt co-supervised the project, supporting Chung-En to use O2 Motion’s unique mobility data products, which the Data and Insights team helped produce from raw cell-tower connections. Guided by the expertise of Professor Elsa Arcaute, from the University College London’s Centre for Advanced Analysis, the team used the mobility data to address limitations of static area classifications such as the ONS Output Area Classification, which is unable to represent how a place’s “character” changes over time.
The analysis highlighted how areas like central Ealing show a different temporal pattern to its immediate surroundings, despite having similar land use. Making use of these temporal differences can help distinguish between parts of the city. For example, Farringdon and Liverpool Streets are both employment hubs during weekdays, but outside working hours, their patterns diverge. Liverpool Street transitions into a hub of nighttime activity while Farringdon’s streets are left empty until the next commuter rush.[OT1.1] The methodology Chung-En developed could help distinguish between these areas, providing more realistic area classifications for cities.
O2 Motion’s data poses its own challenges, like how to translate five-minute mobility counts into features that can be used alongside other traditional datasets. By feeding this data into a transformer model (the deep-learning architecture behind large-language models) Chung-En found this AI technology can successfully condense mobility counts into a lower-dimension representation that’s better suited to analysis. From there areas were classified by grouping together zones with similar mobility and POI characteristics.
Since 2021, Cristobal has co-supervised three Geographic Data Services masters projects, which have all demonstrated the value of applied academic collaboration and of the rich data sources we work with. The work treats cities as dynamic systems and tests methods that combine movement patterns with place context. The outcome is classifications that better reflect how places function across time.
O2 Motion’s data poses its own challenges, like how to translate five-minute mobility counts into features that can be used alongside other traditional datasets. By feeding this data into a transformer model (the deep-learning architecture behind large-language models) Chung-En found this AI technology can successfully condense mobility counts into a lower-dimension representation that’s better suited to analysis. From there areas were classified by grouping together zones with similar mobility and POI characteristics.
Since 2021, Cristobal has co-supervised three Geographic Data Services masters projects, which have all demonstrated the value of applied academic collaboration and of the rich data sources we work with. The work treats cities as dynamic systems and tests methods that combine movement patterns with place context. The outcome is classifications that better reflect how places function across time.
Supervising masters students as part of the Geographic Data Service masters programme has been a highlight of the past few year’s. I’m always impressed by the new ways students find to extract meaning from the mobility data feeds we work with.