Predicting Poverty in African Cities Using Open Source Remotely Sensed Data

Géraldine Duthé, Institut National d'Études Démographiques (INED)
Basile Rousse, Institut national d'études démographiques (INED)
Hervé Bassinga, Institut Supérieur des Sciences de la Population (ISSP) / Université de Ouagadougou
Valérie Golaz, Institut National d'Études Démographiques (INED)
Arlette Simo Fotso, French Institute for Demographic Studies (INED)

Due to a lack of regular statistics, little is known about socioeconomic and environmental dynamics of African cities. Recent advances in spatial imaging technologies, along with their growing accessibility, allow for the acquisition of high-resolution temporal and spatial data worldwide and Open source remotely sensed data provide a large variety of measures for many uses. In this study, we aim to explore the relevance of such kind of sources in order to predict the socioeconomic and environmental characteristics of neighborhoods in African cities. To do so, we use buildings detection provided by Google Open Buildings: models have been trained using deep learning to detect buildings from remote sensing images. Confidence scores are also provided for each detection. The higher the score, the more confident the model in its prediction to distinguish buildings. We assume that this score, in urban cities, can help to identify neighborhoods with no clear urban planning, informal housing and poor socioeconomic conditions. We first use census data to explore the correlation at the neighborhood level in Antananarivo and Ouagadougou. We then use data from Demographic Health Surveys in order to test this association on a larger sample of African cities.

Keywords: Internal Migration and Urbanization, Big data, Population, Environment, and Climate Change

See extended abstract.