Characterising Urban Deprivation through Earth Observation: Linking Physical Urban Form to Socioeconomic Conditions in Zambia

Deprived urban areas (e.g., slums and informal settlements), are characterized by substandard housing, inadequate services, and insecure tenure. They represent the physical manifestation of socioeconomic inequalities across rapidly urbanizing in Low- and Middle-Income Countries (LMICs). Populations residing in these areas face compounding challenges including elevated exposure to climate and environmental hazards (e.g., extreme heat). Yet, these communities are often underrepresented in official censuses, limiting efforts to identify and reach those most in need. Earth Observation (EO) and Machine Learning (ML) offer potential to address this gap, yet current mapping approaches produce mostly binary slum/non-slum classifications that obscure the continuous, multidimensional nature of deprivation.

This research develops a morphology-based framework for characterising urban deprivation in LMICs, using Zambia as a primary case study. Rather than training supervised models on binary slum boundaries, we leverage EO-derived urban elements including building footprints, heights, street network characteristics, and spatial arrangement patterns to compute a set of morphometrics at fine spatial resolution. Applying unsupervised ML techniques, we identify distinct morphological signatures across urban areas. To assess whether and how these signatures relate to deprivation, we integrate household-level data from accurately (~3m) geo-coded urban household surveys in Zambia in 2023 with EO imagery to examine associations between physical urban form and non-physical dimensions of deprivation, such as service access and socioeconomic status. Preliminary results will highlight which morphometrics demonstrate robust associations with socioeconomic indicators and how these relationships may vary across different urban contexts, as well as the rural-urban continuum.

The framework responds to the challenge of transforming globally available EO data into locally actionable information. By producing human-interpretable morphological characterizations rather than abstract deep learning features, the approach offers greater transferability across diverse urban settings and facilitates co-creation with local stakeholders who can validate whether outputs align with their understanding of deprivation patterns on the ground.