Putting satellite data to work on identifying informal settlements in South Africa

A collaboration between Zindi, Amazon Web Services and the South African National Space Agency has yielded a machine learning model that can identify informal settlements from satellite imagery. This tool will help the South African government with planning, providing essential services, and preventing crime in underserved communities across the country.

The image on the left shows the type of satellite data used in the competition; the image on the right shows the models’ pediction of informal settlements (blue). Image credit: Zindi.

“The idea was to get data scientists to work on a potential solution that SANSA could use to optimise our mapping processes,” says Managing Director of Earth Observations at SANSA, Andiswa Mlisa. This model will assist SANSA in mapping of informal settlements, a task that SANSA currently undertakes manually.

Employing the experts

In June 2020, Zindi hosted a hackathon using SANSA data and AWS virtual machines. We invited more than 150 of our best users — Zindians with proven computer vision expertise and success in previous competitions — to take part in this prestigious weekend-long event. They were competing for $1000 USD and a chance to make the world a better place, using virtual machines sponsored by AWS for the computing power needed for machine learning engineering.

Data scientists from 34 countries used SPOT satellite image data provided by SANSA to create the models. Satellite images with manually labelled informal settlements around Johannesburg in Gauteng were used as training data for modelling, and data scientists were challenged to create models that can find informal settlements in KwaZulu-Natal.

After 60 hours of competition, Raphael Kiminya from Kenya came out on top with a model that managed to predict informal settlements that human labellers had missed. Kiminya is a freelance data scientist who believes that AI should be a force for social good in the world.

“The genuine beauty of AI lies in its capacity to touch all facets of our lives,” he says. “There is so much yet to achieve in healthcare, education, governance, and security to name a few. The only limitation is our imagination and how much we care to ask the right questions. I appreciate Zindi for giving me a platform to apply my skill set to positively affect my community.”

Putting data to work

Zindi and SANSA are working together to put these models into practice.

“The models that we have created will provide a kind of heatmap, with different probabilities indicating where an informal settlement is likely to be,” says Zindi CEO, Celina Lee. “What’s nice is that the model can pick up on what the human eye might just scan right over and not notice.”

She says that it was exciting to work with SANSA on this project as it unlocked opportunities for many African data scientists to showcase their talents. This hackathon also illustrates the wealth of data that SANSA has to offer data scientists on the continent.

Read more:

Zindi is a competition platform hosting a community of data scientists dedicated to solving Africa’s toughest challenges through machine learning & AI