Smart but Unequal?

Sensor Infrastructure in the Smart City

Rachel Franklin (@rsfrankl)

Newcastle University | The Alan Turing Institute

2025-02-25

How do emerging technologies intersect with the spatial demography of cities to exacerbate, reproduce, and generate inequalities across areas or groups?

In particular

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

A talk in three parts:

  1. Why care about sensors and sensor networks?

  2. How to conceptualise and measure coverage?

  3. How to support equitable decision-making around sensor placement?

Part 1: Why care about sensors?

1. They’re everywhere

2. Increasingly they’re essential urban infrastructure

  • Policy decisions are made from sensor measurements
  • Sensors and sensor networks help cities run

Perennial challenges of infrastructure

  • Cost
  • Maintenance
  • Engineering and planning
  • Service levels
  • Equity

3. We humans are funny about infrastructure and equity

New forms of infrastructure, same equity challenges

  • Digital and internet divides
  • Energy transitions and net zero
  • Smart city infrastructures (sensors, networks, data, etc.)

“New technologies have a tendency to polarise and divide at many levels” (Batty et al., 2012)

“smart cities are also internally differentiated … they are geographically uneven at a variety of scales. Whatever it means for a city to be ‘smart’, it is also readily apparent that not all spaces of the city will be equally smart, meaning that smart cities will privilege some places, people and activities over others.” (Shelton et al. 2015)

“Geographies of codified knowledge have always been characterized by stark core–periphery patterns, with some parts of the world at the centre of global voice and representation and many others invisible or unheard.” (Graham et al. 2014)

“urban futures anticipated by urban big data assemblages are highly uneven, data and algorithms cannot divest themselves of urban inequalities and the persistence of their geographies” (Leszczynski, 2016)

4. Even with good intentions we miss people and places

Part 2: Conceptualising and measuring coverage

A quick note on sensors and surveillance

When is surveillance a good thing?

How we conceptualise sensor “deserts”


Areas of the city:

  • not represented by sensors
  • for which knowledge is not produced
  • where modelling uncertainty is higher

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

Two optimization approaches:

  • 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.

  • Define the “coverage” of output area 𝑖 due to sensor configuration 𝑗

  • Place sensors to maximize the weighted sum of output area coverages across the city

  • Sensors located at output area centroids

    • Essentially means every selected output area is covered by a sensor, as well as fractions of surrounding areas

Figuring out who’s in the gaps

Coverage for older residents (>65)

Coverage for place-of-work population

Tradeoffs

Coverage for older residents (>65)

Coverage for place-of-work population

Managing coverage tradeoffs

Making good decisions is hard

Making good decisions is hard

Making good decisions is hard

Can we identify “compromise” networks that may not be best at coverage of any one sub-group, but are satisfactory across all?

Maximizing Coverage Across Groups

Multi-Objective Optimization

Maximizing Coverage Across Groups

Multi-Objective Optimization

How Does Our Approach Compare to an Actual Existing Network?

146 Urban Observatory sensors, located in 55 output areas

Part 3: Supporting equitable decision-making

Developing tools to help create networks that all people and places can benefit equally from

Supporting Decision-Making (and planning!)

Supporting Decision-Making (and planning!)

Supporting Decision-Making (and planning!)

Wrapping up

  • Important: What is the purpose of a given sensor network and what might equitable coverage look like?
    • One size may not fit all
  • Valuable role for optimization tools and decision support interfaces to help work through trade-offs and communicate outcomes.
    • Transparency of decision-making
  • Need for clarity around sensor purpose, surveillance, and bias.
  • Potential for incorporating mobility and time-space trajectories into thinking about exposure and coverage.

Final (bigger picture) takeaways


No fancy data here and well-known methods. But should make us think:

  • how are our data produced?

  • how can we do better?

Final (bigger picture) takeaways

  • Multi-disciplinarity
  • Lengthy projects
  • The data-equity interface
    • It’s not always about data—with, e.g., sensors, it’s partly about location decisions
    • Easy for inequities to be hard-wired into infrastructures
  • Spatial inequality as an important (the important?) lens
    • New infrastructures…same old challenges
  • Bridging the research-policy divide

The End.