Google Deploys AI Technology To Document Unmapped Structures In Africa

Google has integrated millions of previously unmapped buildings across Africa into its Maps service, following a significant artificial intelligence project aimed at bridging digital infrastructure gaps on the continent.

Abdoulaye Diack, Programme Manager for Google Research, explained that the initiative involved using satellite imagery and machine learning to identify structures in underserved and remote areas that had never been formally documented.

“When we started, we noticed that many neighbourhoods in Africa weren’t even visible in Google Maps. Our project helped fix that,” Diack said.

“Almost half of the buildings you see on Google Maps in Africa came from our initiative.”

The project utilized European Sentinel satellite data, refreshed every five days, to train models capable of detecting building footprints. To date, more than 150 million buildings have been mapped across Africa, including informal settlements and rural homes.

Diack highlighted that the AI model went beyond just recognizing rooftops—it was trained to estimate building heights by analyzing shadow lengths in satellite images alongside the time of day the images were captured.

“We were able to detect height for over 1.8 billion buildings globally. That wasn’t the original goal, but it was one of the most fascinating outcomes,” he said, noting that this data now supports urban planners, emergency responders, and economists in understanding the layout of African cities.

Starting in Ghana, the project has since expanded worldwide, with AI tools developed for Africa being applied in other regions.

Diack emphasized that data scarcity in Africa was a major driver of innovation: “Because high-resolution images were lacking, we had to use lower-quality satellite data, which pushed us to innovate on model efficiency.”

Beyond mapping infrastructure, Google Research has applied AI to tackle local challenges. One such project, PlantVillage, helps farmers detect diseases in cassava plants via smartphone cameras. Launched in collaboration with Makerere University in Uganda, it involved collecting thousands of images of healthy and diseased leaves to train the model.

“The AI can detect issues before they’re visible to the human eye,” Diack said, describing the data collection process as “simple but critical,” often requiring on-the-ground efforts: “We send people to farms to take pictures of plants, and over time, the model learns by comparing patterns.”

In language technology, Google is building datasets for under-represented African languages through partnerships with organizations like Masakhane and various universities. For speech recognition, participants record themselves speaking native languages or dialects, helping models grasp local languages and regional accents.

A crucial part of Google’s African strategy is developing AI talent. The company previously ran an AI residency program that trained early-career researchers over 18 months.

“People with mathematics and physics backgrounds tend to do well,” Diack said. “We work with institutions like the African Institute for Mathematical Sciences and others to identify and support talent.”

Google has also partnered with groups such as AfricaToML to equip software engineers with AI skills, especially in countries like Nigeria.

On what Africa needs to fully adopt AI, Diack emphasized the importance of local engagement and data accessibility.

“We need more grassroots efforts that train and build communities. And we must invest in datasets that reflect African languages and realities. If AI tools can’t understand you, they can’t serve you,” he concluded.