← Past projects
São Tomé and Príncipe · 2022
Where the poorest households live
Predicted poverty for every populated 110 m cell on both islands. We built the predictions from satellite imagery and building footprints, including the colour of roofs, to show where to enrol households when social safety net programmes expand.
Predicted poverty, fifths of households
Cells are ranked from poorest to least poor and split so that each class holds a fifth of the country's households. The district map uses the same breaks.
How these maps were made
We trained a machine learning model on proxy means test (PMT) scores from the national social registry. The predictors come from very high resolution satellite imagery: building footprints, building size, roof colour, vegetation, roads and distance to the capital. We then predicted a PMT score for every populated cell, and combined it with estimated household counts to locate the poorest 20% and 40% of households.
6,777Populated 110 m cells
44,348Estimated households
Client: Government of São Tomé and Príncipe & the World Bank. Read the paper.