Title: Characterising urban environments in Sub-Saharan Africa with satellite imagery and unsupervised deep learning

Abstract: Sub-Saharan Africa and other developing regions have urbanized extensively, leading to complex urban features with varying presence and types of roads, buildings and vegetation. We use a novel hierarchical deep learning framework to characterize multidimensional urban environments in multiple cities. Application of the model to Accra, Dakar, and Dar es Salaam revealed analogous patterns of building density, roads and vegetation, such as dense settlements within the metropolitan boundary and a peri-urban intermix of natural and built environment, which we attribute to shared colonial legacies and recent growth trajectories. Kigali, with its mountainous geography and post-colonial expansion, exhibited unique urban characteristics including a sparser urban core and significant wildland-urban intermix. Our results demonstrate that unlabeled satellite images with unsupervised deep learning enables near real-time urban monitoring, particularly in regions where traditional data are scarce.

Bio: Dr A Barbara Metzler is a Research Associate with the Science of Cities and Regions group at the Alan Turing Institute. As part of her research, she is leveraging deep learning models and satellite imagery to study urban areas. During her PhD from Imperial College London, she used unsupervised/self-supervised deep learning techniques together with high-resolutions satellite images to classify urban areas in Sub-Saharan Africa for applications in Public Health.