Computer scientists have used satellite imagery and machine-learning techniques to make detailed maps of regions where poverty is common.

Neal Jean and his colleagues at Stanford University in California focused on Nigeria, Tanzania, Uganda, Malawi and Rwanda and combined various data sets, including daytime images that identify features such as paved roads and metal roofs, to estimate local household consumption and income. When identifying areas where incomes are below the international poverty line of US$1.90 per person per day, the team's algorithm outperforms night-light maps (an alternative but limited indicator of economic activity). It also probes hard-to-reach areas, including regions not accessed by household surveys — such as those conducted by the World Bank — which are costly and infrequently conducted.

The method could prove useful for targeting social programmes and determining when and where they fail.

Science 353, 790–794 (2016)