Using remote sensing and machine learning to explore the spatial patterning of poverty and inequality

This interdisciplinary project will focus on studying how we can improve our understanding of the spatial patterning of poverty and inequality across the UK and possibly internationally (depending on the student’s interests/skills and data availability). You will do this by combining existing GIS and administrative data resources on area-level deprivation with novel remote sensing and machine learning/artificial intelligence (AI) approaches. The aim is to better understand how poverty and inequality manifest themselves in the urban environment and how they have changed over time. The remote sensing component will also allow us to have more synoptic, ‘big-picture’ views of areas considering regional and country-level scales. How we measure inequality at this scale will include examining inequalities in factors such as urban sprawl, housing density and access to green space and how they can complement in situ measurements of physical, environmental and socioeconomic variables1.

The UK is a highly geographically unequal country, with measures such as health, poverty, education, employment and wealth showing some of the most distinct differences across our shores, generalised as a ‘North-South’ divide2,3. These measures can be seen as both outcomes and predictors, as well as interacting with each other. Essentially these inequalities represent unequal social and economic experiences and exposures, resulting in detrimental effects on health, wellbeing, social mobility and productivity4. While we have some good evidence about patterns of geographic inequality, there remains scope to better measure and understand the true nature of inequality in the country with a much ‘finer resolution’ and how these inequalities are changing in the rapidly developing urban landscapes. Traditional mapping methods and administrative data struggle to keep up with such rapid changes and therefore limit the evidence for supporting successful and equitable and sustainable ‘smart city’ development. Satellite and remote sensing technology, combined with machine learning and spatial statistics, offer new methods and answers to old problems of how to manage sustainable and equitable urban development and meet the ever-increasing demand from city-based populations. A strong motivation for introducing satellite images into the measurement of inequality, over-and-above the possibility to realise new dimensions of inequality not possible with administrative and ‘on-the-ground’ data, is that we have entered a new era of freely available satellite data (e.g. the ESA Sentinel constellation missions). We are experiencing a rapid growth of activities in the Space industry and the Earth Observation sector and these opportunities do not just support commercial activities, but also provide more efficient tools to the environmental management community and socio-economic research.

Policy Relevance
If we are to realise the UK government’s current ‘levelling up’ focus for example, better evidence of the nature and causes of, and solutions to, UK regional inequalities are needed. This project will therefore provide robust evidence using novel data linkage of existing, siloed data sources combining satellite imagery, geospatial and administrative data to highlight the importance of where we live on patterns of poverty and inequality and how these are changing over time.

Supervisory Team
The supervisory team for this project includes expertise in: assessing inequalities and health inequalities using existing administrative and cohort data (Tony Robertson); remote sensing, machine learning and AI (Armando Marino); and spatial and spatio-temporal statistical modelling (Craig Anderson). You will be based in the Division of Biological and Environmental Sciences (BES) at the University of Stirling alongside Dr Robertson and Dr Marino. Researchers in BES have a unique focus on the relationships between people and the environment. Alongside your work and supervisory team in Stirling, you will be supported by Dr Anderson from the School of Mathematics and Statistics, University of Glasgow and will be encouraged to join and engage with research groups and opportunities across both Universities and the IAPETUS-II scheme.

Click on an image to expand

Image Captions

Dhaka form above, near Baridhara. Attribution: Mohammad Tauheed (no changes made) https://www.flickr.com/photos/12342805@N00/49263301482


In terms of methodologies this project will focus on:

1) AI classification of urban areas: The work with AI and remote sensing aims to advance the classification and characterisation of urban environment into classes that have socioeconomic meaning in the context of studying inequality in the country under analysis. Research has been carried out in detecting urban areas5,6, for instance focusing on informal settlements7, however the problem has been generally approached from a land-use prospective. In this project, we want to use AI to have a more holistic approach to urban classification which is strongly linked to socioeconomic drivers of inequality. Fieldwork will also be introduced to the project here to validate and finalise our classifications.

One of the expected challenges that you will need to solve is the fusion of satellite data from different sensors, together with other administrative and GIS datasets. You will approach this by considering Deep Learning8. To combat the paucity of training data we will using Transfer Learning (e.g. where models previously built for optical sensors are adapted and used for radar sensors).

2) Spatio-temporal modelling: Once the classification maps of urban areas are obtained, you will evaluate the spatio-temporal patterns of inequality using novel, quantitative statistical analyses9. This will incorporate spatial modelling, and more specifically, will focus on physical-spatial relationships via geographic coordinates and also practical-spatial relationships via the likes of travelling times and access to services. The relationships between our new inequality measure(s) and outcomes such as population health measures could then be explored.

Project Timeline

Year 1

• Explore the literature on geographical deprivation and inequalities
• Apply for ethical approve and data access
• Data cleaning
• Training for methodological skills gaps
• Process data and begin to develop the machine learning models
• Begin write-up of introduction chapter(s)

Year 2

Refine and test the machine learning models
• Training for additional skills gaps (methodological and generic)
• Complete introduction chapter(s) and begin methods chapter(s)
• Fieldwork to compare the modelling data to ‘on-the-ground’ data.
• Submit journal article on classification of urban areas using AI

Year 3

• ‘Full-scale’ modelling investigating patterns of inequality
• Complete write-up of methods chapter(s) and begin write-up of results chapters
• Final analytical work on full-scale modelling
• Submit journal article on assessment of urban areas on a country scale using AI
• Submit journal article on inequality patterns using new classification scheme linked to health/social/economic outcomes

Year 3.5

• Complete write-up of results chapters and discussion chapter
• Produce final version of the completed thesis

& Skills

Training and skills development will be aligned to the student’s skills and experience and agreed between the student and the supervisory team. We envision face-to-face, online and hybrid training will be accessed in areas related research methods in machine learning, remote sensing/satellite imagery, social and environmental deprivation/inequalities and spatial statistics. Personal, professional and further research skill development will be identified through a skills gap analysis in the first 3 months of the PhD and these needs addressed through attendance at the University of Stirling’s Institute for Advanced Studies activities including induction, skills development courses, conferences and other development opportunities as identified and required. You will join and become an active member of the Environmental Sustainability and Human Health and the Earth and Planetary Observation Sciences research groups at the University of Stirling and the Environmental Statistics research group at the University of Glasgow. Membership will allow for you to access unique training, presentation and collaborative activities across the range of disciplines covered by these groups.

References & further reading

1. Miller, R. B. & Small, C. Cities from space: potential applications of remote sensing in urban environmental research and policy. Environmental Science & Policy 6, 129–137 (2003).
2. Dorling, D. Persistent north-south divides. in In: Coe, N.M., Jones, A. (eds), The Economic Geography of the UK 12–28 (Sage Publications, 2010).
3. Office for National Statistics. What are the regional differences in income and productivity? https://www.ons.gov.uk/visualisations/dvc1370/ (2021).
4. Agrawal, S. & Phillips, D. Catching up or falling behind? Geographical inequalities in the UK and how they have changed in recent years. https://ifs.org.uk/inequality/geographical-inequalities-in-the-uk/ (2020).
5. Maktav, D., Erbek, F. S. & Jürgens, C. Remote sensing of urban areas. International Journal of Remote Sensing 26, 655–659 (2005).
6. Xia, N., Cheng, L. & Li, M. Mapping Urban Areas Using a Combination of Remote Sensing and Geolocation Data. Remote Sensing 11, 1470 (2019).
7. N. Mboga, C. Persello, J. R. Bergado, & A. Stein. Detection of informal settlements from VHR satellite images using convolutional neural networks. in 2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS) 5169–5172 (2017). doi:10.1109/IGARSS.2017.8128166.
8. Suel, E., Bhatt, S., Brauer, M., Flaxman, S. & Ezzati, M. Multimodal deep learning from satellite and street-level imagery for measuring income, overcrowding, and environmental deprivation in urban areas. Remote Sensing of Environment 257, 112339 (2021).
9. Anderson, C., Lee, D. & Dean, N. Identifying clusters in Bayesian disease mapping. Biostatistics 15, 457–469 (2014).

Apply Now