"Using high-resolution satellite images to improve public policies-
Example from an urban social protection program in Mozambique
The rapid expansion of publicly available high-resolution satellite images holds great promises for improved targeting and evaluation of policy interventions, especially in urban areas of developing countries. This note describes how Google Earth (GE) images have been successfully used to guide an urban social assistance program in Tete, Mozambique. In particular, we extract information from GE on housing types, roof type, building size, and overall orderliness of a neighborhood. In a second step we combine this (and other GIS variables) with geo-referenced survey data in order to predict eligibility to a public works program.
We estimate a welfare proxy (PMT scores) using ordinary least squares and random forest and find correlations between predicted and actual values of 0.6 and 0.9 respectively, allowing for detailed mapping of average welfare. The approach could be generalized to other areas of interest such as population, need for public services as education and infrastructure, or other social outcomes. Further, the continuous flow of new images also allows for frequent updates to ongoing programs and impact evaluations using either historic data or future data. In time, datasets such as this could be used to train artificial intelligence algorithms to tag new images automatically.
Contact person: Neda Trifkovic