Monitoring Urban Growth in Informal Settlements Using Deep Learning: A Case Study of Khayelitsha, South Africa

Rapid urbanisation has accelerated the expansion of informal settlements in many African cities, creating significant challenges for urban planning, service delivery, and disaster risk management. Traditional mapping approaches are often unable to keep pace with rapid settlement growth, resulting in outdated spatial information. This study evaluated the use of a deep learning-based geospatial workflow to quantify informal settlement expansion and changes in environmental hazard exposure in Khayelitsha, Cape Town, between 2013 and 2022. Very high-resolution (0.08 m) orthophotographs supplied by the City of Cape Town were analysed using the Deepness plugin in QGIS with a pre-trained XUnet building segmentation model. Model performance was validated against manually digitised reference data using Precision, Recall and F1-score metrics before being applied to multi-temporal dwelling extraction. Settlement growth was quantified using dwelling counts and Nearest Neighbour Index (NNI) analysis, while changes in environmental vulnerability were assessed through proximity analysis to flood-susceptible wetland areas. The deep learning workflow achieved an F1-score of 74.8% (Precision = 66.5%, Recall = 85.6%), demonstrating its suitability for automated dwelling extraction in a complex informal settlement environment. Between 2013 and 2022, the number of informal dwellings increased by 282.4%, representing an additional 2,895 structures. NNI analysis revealed that growth occurred primarily through outward spatial expansion rather than densification, with the index increasing from 0.58 to 0.94. This expansion resulted in a measurable increase in environmental vulnerability, with the average dwelling located 45.81 m closer to flood-susceptible areas by 2022.