The sensor desert quandary: What does it mean (not) to count in the smart city?

By Cait Robinson, University of Liverpool, and Rachel S. Franklin, Newcastle University

As COVID-19 has brought the need for improved digital infrastructures into sharp focus, smart cities have been pitched as a solution. Smart cities collect and stream real time data at a high resolution, with the aim of improving services for citizens whilst reducing government costs. Sensor technologies are central to smart cities. Smart city sensors can measure a whole host of urban characteristics including air quality, temperature, traffic, footfall and noise. Subsequently, sensor networks are big business and during 2020 cities are estimated to have spent $20 billion on sensor technologies.

Yet in many national contexts, expensive and expansive smart city agendas have emerged during an era of post-2008 crisis austerity and growing global and regional inequality. Despite aiming to produce new data and knowledge about cities, sensor technologies actually have the potential to reproduce well-documented social injustices. However, one specific inequality, which we have highlighted in a recent paper in Transactions of the Institute of British Geographers, occurs when sensors create gaps in understanding about specific urban populations, what we term sensor deserts. Based on our ongoing research, this blogpost defines and illustrates sensor deserts in smart cities.

Spatial inequality in the smart city

There are many different definitions of what it means for a city to be smart, and in turn of who or what the smart city might be for. In the case of flagship projects driven by Big Tech it is hard to imagine how they could benefit anyone but the ultra-elite. Meanwhile, in ‘actually-existing smart cities’, such as Newcastle’s Urban Observatory or Future City Glasgow, most ‘smart’ interventions are not built from nothing, but must be awkwardly integrated into existing urban structures. Enacted by public and private sector actors, these projects must balance achieving economic growth with solving social problems.

As smart city agendas have progressed, increasing evidence has emerged of the inequalities embedded within them. Smart technologies and infrastructures have a tendency to reproduce well-documented historically and geographically embedded patterns of privilege, whilst also revealing new geographies of inequality. For example, in the case of India’s recent 100 smart cities initiative, democratic governance processes have been weakened. As economic growth is prioritised over the wellbeing and participation of ordinary citizens, this has led to infrastructure development that benefits the wealthy and further marginalise the poor. Sensor technologies and networks are likely to be no exception.

Defining sensor deserts

Although sensors might appear to be spread evenly throughout a city, in fact, some neighbourhoods and places will be covered while others are not. Sensor deserts then are those areas within a smart city that are lacking in coverage, and therefore the investment, representation and legitimacy associated with smart city agendas. Whilst critics have voiced genuine concerns about surveillance in smart cities, emphasis upon a lack of coverage allows us to consider what might be thought of as the opposite of over-surveillance, asking instead what it means (not) to count in the smart city?

Sensor deserts can be seen in two ways: in the physical sensor infrastructures and the data produced. Whilst many components of the smart city are hidden from view in offices or the cloud, sensor infrastructures have a visible, material presence in cities, attached to lampposts or installed on kerbsides. Their deployment is indicative of investment priorities, and therefore which people and places are represented and legitimised by smart city efforts. Decisions about the placement of sensor infrastructures also shape the gaps and subsequent biases in the data that are collected. Crucially, owing to the cost of installing and maintaining sensors, policy-makers cannot provide universal coverage of all areas and are necessarily selective in terms of sensor placement. As such deserts by data—those places that fall into the gaps in sensor data collection—also emerge.

Deserts in data are by no means a new phenomenon and there has long been interest in the ways in which people and places are made (in)visible by statistics. The US census undercount is a typical example of this, with the worst undercount in three decades forecast for the 2020 census. Four million people risk being missed from estimates, particularly affecting African American (-3.68%) and Hispanic (-3.57%) populations (Urban Institute, 2019). Despite claims about ubiquity of sensing infrastructures, smart cities have created a ‘data deluge’ in some areas whilst ‘data deserts’ continue to exist.

Sensor coverage in practice

In practice, sensor networks must often fulfil a variety of functions (see Table). The placement of sensors is shaped by the need to balance competing demands from different stakeholders and communities with technical practicalities – with implications for what constitutes good coverage. A market-based approach is likely to result in deployment in more lucrative neighbourhoods. Meanwhile placement can also be politically motivated, with civic leaders knowing which areas need to be “bought off”. Those concerned with social justice are likely to locate sensors equitably, either in areas most exposed to urban challenges or amongst vulnerable populations. For the purpose of representation, even coverage across the city might be most desirable. Data from sensors if also used as an input into models that estimate values for areas without sensor coverage. Subsequently, locations that are likely to provide better model inputs may be prioritised. Meanwhile, the complexity of the urban environment in which sensors are placed determines whether a sensor is representative. Each of these aspects has implications for what a sensor captures or emphasises.

Types of sensor desert, author provided.

Average distance to nearest air quality sensor in Newcastle-upon-Tyne, based on the Urban Observatory network (UO, 2020), author provided.

This complexity translates into sensor networks that are often highly context specific. For example, in the case of Newcastle-upon-Tyne’s Urban Observatory (UO) – an ambitious £2 million smart city that provides access to the largest public, real-time dataset in the UK – decisions about sensor placement are made on a relatively ad-hoc basis, in response to local government priorities, funding pots targeted at specific issues, or requests from local community groups. The network must fulfil a variety of functions and therefore provides differential coverage across vulnerable populations in the Local Authority. For example, analysis of the average distance to the nearest air quality sensor illustrates how private renters and ethnic minority groups are typically closer to a sensor, but social renters and those with poor health or a disability are further away (see Figure). However, as a public network committed to open data the UO represents best practice compared to smart cities where data is inaccessible, or behind a paywall.

With the global sensor market forecast to grow by a further 24% over the next five years, sensors are likely to become an increasingly familiar sight within our cities. Considering equity in sensor placement—or, who counts— is crucial as sensor networks expand, and increasingly inform decision-making. Deserts in coverage should be on the agenda of urban policymakers concerned with implementing and evaluating these new sensor technologies.

About the authors: Cait Robinson is a Lecturer in Urban Analytics in the Geographic Data Science Laboratory and Department of Geography and Planning at University of Liverpool. As a quantitative human geographer, her research explores the causes and consequences of spatial inequality, with a particular interest in energy and infrastructure. Rachel Franklin is Professor of Geographical Analysis in the Centre for Urban and Regional Development Studies (CURDS) and the School of Geography, Politics and Sociology at Newcastle University. She is a population geographer whose research focuses on drivers and impacts of demographic change, including most recently the spatial demography of smart cities, and local and regional population loss.

The research in this paper is part of the Spatial Inequality in the Smart City project funded by the Alan Turing Institute project #R-NEW-001. The datasets and code underpinning the descriptive analysis in the paper are available here: https://github.com/CaitHRobinson/SpatialInequalityintheSmartCity