Data

• Historical – maps, diaries, inventories, censuses, surveys, archeological sites

• Big and smart – sensors, social media, internet, mobility, street view

• Government and administrative – census, medical, vital records, administrative

• Linked – joining or nesting information across locations, time, or individual

• From space and from the air – satellite imagery, aerial photography, drones, LiDAR

Methods

• “Traditional” methods to ascertain relationships and associations across variables and places, collapse data dimensionality, compare groups and outcomes

• Data science and visualisation

• Machine learning

• AI: computing and analytical innovations that facilitate data discovery and manipulation, text analysis, feature extraction, data creation, and analytics at scale
  
  
  

As social scientists and health researchers we can occasionally get distracted by the shiny data and methods objects. The attraction of the novel.

The “red herrings”, to me, are the ways in which data and methods distract us from the real problem of improving wellbeing and health, and reducing spatial inequalities.

There’s no such thing as AI, as in “let’s use AI for this”

Also: the problem at hand should determine the method.

Exciting and cool, but also potentially problematic when it comes to issues of bias, representation, and privacy. We should be more open about this.

Most novel data sources still rely on traditional data for validation. We need to be careful not to throw the baby out with the bathwater.

A match made in heaven and so much potential. But also so much temptation to put the cart before the horse.

Thinking about how emerging technologies intersect with the spatial demography of cities to exacerbate, reproduce, and generate inequalities across areas or groups.

Especially placement of sensors in the urban landscape

• Sensor coverage and “sensor deserts”
• Who’s in the “gaps”
• Equitable decision-making

Even with good intentions infrastructures miss people and places

The big idea

How can we support informed and equitable decision-making around sensor placement, especially what criteria ideal networks might satisfy and the inevitable trade-offs involved?

We’ve got good optimisation algorithms for this

• What is the ”best” allocation of n sensors, given a particular goal?

  • Single objective greedy algorithm: Place sensors one by one and maximize coverage of one sub-group only.
  • Multi-objective genetic algorithm (NSGA2): Generate a spectrum of networks representing the coverage trade-offs between different sub-groups.

• Decision support tools that visualize options and trade-offs

Coverage for older residents (>65)

Coverage for place-of-work population

Making it more user-friendly

No fancy data here and well-known methods

But should make us think:

1. how are our data produced?

2. how can we do better?

Thinking about the ways in which health, climate change, and transportation intersect

Case in point: The London Tube

• The first Tube line opened in 1863 and is still running as part of the Metropolitan line
• 272 stations
• 11 lines, covering 402 kilometres
• More than five million passengers on the busiest days

Future heat on the London Tube

• Only 4 out of 11 London underground lines have air conditioning systems

• The average summer temperature in London is expected to increase by 2.7 degrees Celsius by the 2050s

• The probability of heatwaves could also increase five-fold and they’re expected to occur every other year

• By 2070, the mean maximum air temperature in the UK in August is projected to increase by up to 6 °C in summer compared to 2018

Hot is already here: Average station temperatures in London (2019)

The big idea

How can we estimate current and future heat exposure on the Tube and who (where) is most affected?

Simple (and important) question with complex data requirements

• Travel flows (origins and destinations)

• Who’s travelling? (demographic and health characteristics)

• What’s the temperature on board? (estimated from known station-surface differentials)

Exciting geographical tools

• Synthetic Population Catalyst (SPC)–A synthetic population and its extended AcBM (activity-based modelling) dataset that simulates individual-level travel behaviour (homeplace and workplace), travel mode, person and socio-economic factors to allow us to explore heat vulnerability at the individual level.

• Tube operation timetable–For travel times and route estimation, we use the timetable provided by TfL APIs. Provides accurate estimation whether travellers for each OD-pair will take air-conditioned Tube lines.

• Clim-recal–Estimates weather and heat wave days in the past and future on a daily basis in 2.2 km*2.2 km cells covering the entire UK. Local variation in the dataset is used to estimate heat exposure more accurately.

Preliminary results (a sense of where we’re heading)

• The inequality of heat exposure risk is significant in spatial terms

• Trickiness of estimation but lots of useful data that can be brought to bear

• But also some of this should be being measured directly!

The big idea

We’ve now got the compute, methods, and sensor quality for satellite imagery to be a game-changer for social science and health research and policy making

Introducing the ESRC SDR Imagery Data Service*

1. Imagery innovation–research-ready imagery-based data products, building off and developing innovative computing and AI methods that facilitate efficient automated workflows for measures and indicators, as well as custom-defined geographies and time periods.

2. Data for all–data distribution channels that meet researchers and policymakers where they are, with user-friendly interfaces, familiar file formats, linkage and integration with existing data resources

3. Capability and community–building capacity for understanding and working with imagery and imagery-derived data, growing the user-base and providing thought leadership, and heightening awareness and enthusiasm for the value of imagery

Something that’s not a red herring

AI and novel data as true game changers

The question chooses the method (and the data).

(Not the other way around)

Complex spatial inequality and health challenges require complex approaches¹

• Inter-disciplinarity
• Data integration
• A lot of social science under the hood ²