Stanford University experts have developed an inexpensive and accurate way to predict village-level wealth. Their computer model uses publicly available satellite images to identify exactly which neighborhoods or villages are home to the poorest communities in a given country. The findings were published in Science in July 2016.
Nighttime satellite images give a clear picture of wealth distribution around the world: the less light at night, the poorer the region. However, detailed and accurate local information is needed in order to create policy and distribute aid. This would mean spending millions of dollars to send thousands of survey-takers into these areas.
The new program promises the same information through a machine learning technique (the science of designing computer algorithms that learn from data) called convolutional neural network and high-resolution satellite imagery.
“Using the final model that has been trained on survey data, we can estimate per capita consumption expenditure for any location where we have daytime satellite imagery,” the team said on the study’s dedicated website.
The team tested their model in Nigeria, Tanzania, Uganda, Malawi, and Rwanda—five African countries for which there is reliable information about distribution of poverty.
“Without being told what to look for, our machine-learning algorithm learned to pick out of the imagery many things that are easily recognizable to humans—things like roads, urban areas, and farmland,” said Neal Jean, the study’s lead author.
The program also learned how to distinguish between metal rooftops and those made of grass or mud. The team then used statistical methods to determine the significance of these items to income.
Eventually, the computer model was able to predict average spending by households and average household wealth. The results matched the available data in these five countries. In some cases, it even offered more accurate results than currently available data.