Can We See Poverty from the Sky?
How satellite imagery and machine learning predict poverty across five African countries where surveys fall short.
Article
Between 2000 and 2010, 39 of 59 African countries ran fewer than two economic surveys. Researchers combined freely available daytime and nighttime satellite images with machine learning to estimate poverty where ground data is missing.
A transfer-learning model learns from nighttime light intensity, then reads roads, roofing materials, and farmland in daytime images. It explained 75% of the variation in asset wealth and 55% in household consumption across Nigeria, Tanzania, Uganda, Malawi, and Rwanda, giving policymakers a way to target aid, education, and infrastructure.
Visual story
