AI Case Study
Stanford University’s Department of Earth System Science identified the most poverty-stricken regions in parts of Africa using machine analysis of satellite imagery
Traditionally satellite imagery has been used to identify areas of relative poverty by visualising the density of electric light seen at night time. A team from Stanford University’s Department of Earth System Science used both night and day satellite images to identify signifiers in the daytime pictures.
Combining the image sets helped the algorithm to predict poverty in these regions. Comparison with household survey data from the relevant areas revealed that predictions of poverty were at 81-99% accuracy level compared to night time satellite images alone.
Public And Social Sector
Education And Academia
"Stanford University’s Department of Earth System Science used satellite images of areas in daylight in research in order to ‘fill in the gaps’ of nighttime images alone in identifying the most poverty-stricken regions in parts of Africa. Feeding an algorithm with both night and day satellite images of Rwanda, Nigeria, Uganda, Malawi, and Tanzania, they were able to identify signifiers in the daytime pictures. Combining both image sets, this helped the computer predict poverty in these regions. When compared with survey data obtained from households within them - this led Burke and his team to be able to predict poverty with an 81-99% accuracy as compared to the night time satellite images alone.
AI, in this case, can not only help with the identification of areas most in need of aid, but also help organisations and aid workers on the ground in locations to measure how effective their efforts are in combating poverty."
"Identifying poverty, and the regions that are most in need is a key component in being able to tackle the problem of poverty. Satellite imagery is helping researchers do just this. An abundance of images taken by satellites on a constant stream can lend a hand to identifying global activities that reflect poor and rich regions - areas with a high density of light at night are typically wealthier than those in darkness, with little or no access to electricity over nighttime periods."
Improved capability to predict poverty in an area - 81-99% accuracy compared to previous methodology.