A team of scientists from Stanford University has developed a way of identifying and mapping poverty using satellite technology and machine learning.
Identifying impoverished areas is traditionally a painstaking, labourious and sometimes inaccurate task. Data gathering is usually done via citizen surveys but these have their limitations; they’re not always cost-effective, they’re time-consuming and some regions might not be accessible due to lack of infrastructure or conflict. Late last year, the World Bank released a report that found that 29 countries had no poverty data recorded between 2002 and 2011. Yet proper identification is crucial in order to ensure effective, targeted delivery of aid supplies and services.
A recent study by a team from Stanford University’s School of Earth, Energy and Environmental Sciences may indicate a possible breakthrough. Using satellite imagery and machine learning (a form of artificial intelligence that sees algorithms learn from and adapt to new data), scientists were able to gain insight into poverty levels in Nigeria, Uganda, Tanzania, Rwanda and Malawi.
Comparing Night and Day
In order to get a well-rounded idea of an area’s socio-economic development, researchers relied on both daytime and nighttime images. Nighttime images can indicate a region’s rate of electrification, one signifier of development. These were then matched up against signifiers found in daytime images – such as road infrastructure, rate of urbanisation, type of roof – to determine levels of wealth in a particular region.
The team evaluated results against survey data and found that their methodology was able to predict the distribution of poverty more accurately than existing measures do. Mapping out poverty using this system would help aid organisations tailor their poverty alleviation programmes accordingly and would also be cheap, a boon for non-profits and agencies operating on tight budgets.
A similar approach has been used to identify poorer areas in cities by measuring the amount of trees present in satellite images of particular neighbourhoods. Neighbourhoods with fewer trees tended to be poorer than areas with lots of green. Researchers at Oxford University have also analysed the presence of light in satellite images taken over rural areas to assess levels of poverty.
You can learn more about how satellite technology is being used for social and environmental good via our RESET Special: Drones and Satellites for Good or find out more about Stanford’s model of predicting poverty via their website.