Don't like to read?
According to The Verge, a team of Stanford University researchers has developed a more efficient way to track the poverty of the world. They are also able to see crop conditions on rural farms and illegal deforestation around the world, in hard to reach places. Scientists can do this with the creation of a deep learning algorithm that can recognize signs of poverty through satellite images.
How Does It Work?
Creating the algorithm involved a dual process called transfer learning. In the first step, researchers build the data with a sensory network of images taken during the day and night. This information includes images of five African countries: Malawi, Uganda, Nigeria, Tanzania, and Rwanda. The model was taught, by using deep learning techniques, to anticipate where lights would be at night, in contrast to those during the day, and search for correlations. This helps the satellite to find possible areas of poverty in the African countries.
The model and scientists can distinguish between the areas that should be dark at night, like lakes and dirt roads, and disregard those, versus villages that are well-lit and populated. The images are carefully reviewed by researchers and governments, in efforts to help the poor communities in the five African countries.
The second step of the process involves researchers using a different model called a ridge regression model. Scientists use this model, in the algorithm, to feed the satellite more information. The data includes real survey statistics taken from World Bank Living Standards Measurements Study and Demographic Health Services.
The model does not just learn that 10 houses in a village will most likely mean a certain amount of lights, it cross-checks it with the survey data it has been given. For example, if 10 houses have an income of $1.90 a day, it will now deduce the houses that do not have any survey data hold similar wealth.
The Idea Behind Tracking Poverty by Satellite
Poverty is harder to measure in developing countries. Although household surveys give feedback on economic data, like assets and household wealth, it does not show the economic growth. Neal Jean, the study’s co-author and doctoral candidate at Standford, stated that unfortunately, information is not available for most of the poorer countries because the surveys are expensive to conduct.
The idea is that if we train our models right, they help us predict poverty in areas where we don’t have the surveys, which will help out aid orgs that are working on this issue.
Jean says that transfer learning is the most accurate gauging of the average wealth of villages and household consumption. However, there are limits to this method. An economics researcher at the London School of Economics, Simon Franklin, reported that the system does well when looking at the rural and urban areas and the differences in poverty.
Franklin goes on to explain that the satellite does not measure poverty in cities well. The reason being is that larger cities have both poor and rich areas in close proximity making it harder to use daytime imagery alone.
NASA has recorded night lights since 2012, using a new and more accurate satellite. According to economist Alexi Abrahams from the University of California, NASA’s satellite does not use the older series images, which are being used for the current studies in African countries.
However, Jean points out that another limit to the satellite’s technology is that it is looking, specifically, at African countries. Therefore, the satellite data would not be as accurate if it had to predict poverty in China or India, which are densely populated. Since the method is cheap and easy to scale and the images are from the public domain the scientists say the next step is to work on other countries so they can better map poverty around the world.
By Tracy Blake
Edited by Jeanette Smith
The Verge: Satellite images of Earth help us predict poverty better than ever
Science: Combining satellite imagery and machine learning to predict poverty
The Christian Science Monitor: How satellite images and deep learning can fight global poverty
Los Angeles Times: How to track poverty from space
Image Courtesy of NASA’s Marshall Space Flight Center Flickr Page – Creative Commons License