By Atoyebi Nike
Google has significantly expanded its digital mapping coverage in Africa by adding millions of previously undocumented buildings to Google Maps, thanks to an ambitious artificial intelligence and satellite imagery project aimed at bridging infrastructure data gaps across the continent.
The project, led by Google Research, involved deploying machine learning models to analyze satellite images, particularly in under-mapped, remote, and informal settlements. According to Abdoulaye Diack, Programme Manager for Google Research, nearly half of all buildings currently visible on Google Maps in Africa were mapped through this effort.
“When we started, many African neighborhoods didn’t appear on Maps,” said Diack . “Our work has helped change that reality.”
Using data from European Sentinel satellites, which update every five days, the team trained models to detect building footprints and even estimate building heights, using a method that factors in shadow lengths and sunlight angles. As a result, the AI system has successfully measured the height of over 1.8 billion buildings globally, aiding planners, emergency services, and urban analysts.
Originally launched in Ghana, the mapping initiative has since expanded worldwide, with AI tools developed for Africa now being repurposed for global use.
Diack said that data scarcity in Africa pushed the team to innovate: “With limited access to high-resolution imagery, we had to optimize for efficiency and accuracy with lower-quality satellite data.”
Google Research’s work extends beyond mapping. One of its landmark AI applications, PlantVillage, helps cassava farmers detect crop diseases early using smartphone cameras. Initially developed in partnership with Makerere University in Uganda, the model was trained on thousands of annotated leaf images to detect symptoms before they become visible.
“Sometimes we deploy people to farms just to gather images,” Diack explained. “That kind of ground-level data collection is what makes AI work in real life.”
In the language space, Google is investing in under-represented African languages, collaborating with groups like Masakhane and various African universities. Locals contribute by recording phrases in native languages, helping the AI learn regional accents and dialects for speech recognition.
Diack also highlighted AI talent development as a critical focus. Google has run residency programs for early-career researchers and partnered with organizations such as the African Institute for Mathematical Sciences and AfricaToML to train young Africans, particularly in Nigeria.
“Africa has immense potential in AI. But for it to work, the tools must be designed with our realities in mind,” Diack said. “We need more data reflecting our languages, more inclusive communities, and grassroots engagement.”
