NAIROBI — Artificial intelligence models developed in Western nations frequently fail in agricultural applications across the Global South, unable to recognize local crop types or account for regional infrastructure limits. Scientists and organizations are now building tailored AI systems to address food security and climate change impacts in African nations.
Catherine Nakalembe, an assistant professor at the University of Maryland and Africa program director at NASA Harvest, encountered this problem while mapping crop types in western Kenya. Existing AI systems could not analyze satellite imagery because they did not recognize local crops.
Nakalembe collected her own training data, equipping dozens of volunteers with GoPro cameras mounted on helmets. Her team gathered over 5 million images in two weeks, using them to train facial recognition technology to identify local crops such as maize, beans, and cassava.
Her work now maps cropland, classifies crop types, and estimates yields in Uganda, Kenya, Senegal, and other African countries using machine learning and computer vision. "AI systems from the West are often useless unless adapted for local contexts," Nakalembe said. "They fail to account for the realities of the Global South, including high internet costs, limited bandwidth, and a lack of labeled training data."
Without adaptation, she said, "these systems remain irrelevant, potentially deepening existing inequalities in wealth and access to resources."
Oren Ahoobim, a partner at Dalberg Advisors, said that satellite imagery, video, and other data sources have improved significantly in quality and availability. This translates to more accurate forecasting for farmers, allowing better decisions on planting, fertilizer use, and disease management.
Agriculture provides livelihoods for more than 2 billion people in low and middle-income countries, who face particular vulnerability to climate change impacts that reduce crop yields and incomes. Governments and organizations are pursuing AI applications to address deforestation and food security in rural communities.
Ending hunger is one of the United Nations' 17 sustainable development goals for 2030, but the target is likely to be missed. Approximately 28 percent of the global population, or about 2.3 billion people, are currently classified as moderately or severely food insecure.
Beyond crop mapping, AI is finding other uses. In Brazil's Pará state, the environmental nonprofit Rare uses AI to convert real-time coastal data into WhatsApp voice alerts for fishers and oyster farmers. Microsoft has deployed AI models using bioacoustics for deforestation monitoring.
