Random information #24
Mapping poverty
What is poverty?
How do you define poverty?
Which indicators should be used to assess poverty?
source: pixabay.com
The questions are important and the answers seems rather easy. But are they?
daily income? it depends on the context (cost of life in the country, etc.)
lack of food and level of healthcare are obvious indicators easy to assess on a local scale but not at a regional scale and across different countries.
Creativity is needed to tackle this limitation
Tools: Satellite imagery and machine learning
Roads, farmland, urban areas are elements of human presence. Extracting these elements from satellite images it is possible to predict village-level wealth. Also, areas that are brighter at night are usually more developed. Seems logic, right?
What a resesarch team from Stanford did was to focus on five Arican countries (Nigeria, Uganda, Tanzania, Rwanda and Malawi) and apply a machine learning algorithm that automatically extracted this kind of data from satellite imagery.
"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,"
And now the good part
It worked!
The method had a good performance predicting the distribution of poverty. It even outperformed existing approaches.
It means they found a cheap way to map poverty around the world using only satellite images.
Aid organizations and policymakers now have a new tool to assist them.
Study: Combining satellite imagery and machine learning to predict poverty
I thought it was a good idea to give you an example of something similar. Here are some images from my work.
The second is a satellite image and you can identify the road and the group of houses, just by looking at it. Add all the other things we cannot see but the sensor can and the machine learning methods... et voilà.
satellite image of the previous location. (Sentinel 2B satellite - 10 bands raster stack)
See you tomorrow for more Random information
RMach