Earth observation + morphometrics: towards a systematic understanding of cities in challenging contexts

The ongoing fast and unprecedented urbanisation, strongly affecting cities in the Global South, is casting urban development into challenging dynamics, involving, for example, the construction of new or expansion of existing informal settlements lacking the most basic services. This informal urban development is insufficiently understood due to limited resources, leading to an absence of data to support its urban analysis and, consequently, obstacles to address social equity, sustainability, and resilience through urban planning and design practices. Although the potential of Earth Observation (EO) has widely been recognized for mapping different urban patterns, the surge of big data and data science, along with usually costly high resolution datasets are marginalizing the open science of interpretable urban morphology with the abstract relationship between mapped socioeconomic patterns and image features. In this work, an "EO + Morphometrics" framework is proposed to combine open EO data with open tools for explicit, reliable and consistent measurements of urban form and thus to understand development patterns based on urban morphology. More specifically, a reengineered convolutional neural network is applied to freely available Google Earth imagery to extract building footprints for entire cities. Morphometrics then uses this information to measure hundreds of metrics of the built environment and output homogenous urban form types. This two-step method is applied to a few fast-growing sub-Saharan cities. To test whether specific urban types correspond to socioeconomic patterns, we compare the outcomes with local delineations of informal settlements. Morphological types, characterised by a compact/organic urban fabric, seem to be predominantly contained in the boundaries of informal settlements. We argue that "EO + Morphometrics" paves the way for deriving a generalizable understanding of urban form in challenging contexts. This information can, in turn, be used for further analysis (for example, with socioeconomic data) and inform local planning and interventions.