An innovative early warning system powered by artificial intelligence is set to transform how Africa confronts a silent but deadly threat to its food supply: aflatoxin contamination in maize.
The Aflatoxin Risk Early Warning System (A-EWS) harnesses machine learning and satellite data to predict and map contamination hotspots, providing a powerful new tool in the fight for food safety. The system was developed by the International Institute of Tropical Agriculture (IITA) and its partners through the CGIAR Scaling for Impact Program (S4I) and the CGIAR Sustainable Farming Program (SFP).
According to Barbra Muzata, IITA spokesperson, the A-EWS was showcased at the recent 11th African Grain Trade Summit in Zanzibar, a premier gathering of policymakers and leaders in the grain industry.
Muzata explained that aflatoxin, a potent mycotoxin produced by the Aspergillus flavus fungus, is often described as an “invisible enemy” lurking in the soil. It contaminates staple crops such as maize, groundnuts, and sorghum, leading to severe health problems including stunted growth in children, immune suppression and liver cancer. Economically, aflatoxin contamination costs Africa an estimated $670 million annually in lost grain trade.
“As there is limited information on the spatial distribution of aflatoxin risk at the farm level, understanding this is critical for guiding targeted management interventions,” she said.
Also read: Peanut butter manufacturer fined R500 000
Harnessing Satellite Data and AI to Predict Contamination
The A-EWS fills this information gap by using geospatial artificial intelligence (GeoAI) to forecast pre-harvest contamination risk zones. Its machine learning models were trained on a comprehensive dataset that combines:
- Historical data: 907 unique pre-harvest maize samples from farms in Kenya, Uganda, Malawi, and Tanzania collected between 2009 and 2022.
- Environmental variables: Satellite-derived data on temperature, precipitation, humidity, elevation, and soil properties, key factors influencing fungal growth.
The system classifies aflatoxin risk into three internationally recognised categories:
- Low (<5 ppb)
- Medium (5–20 ppb)
- High (>20 ppb)
Among eight machine learning algorithms tested, the Gradient Boosting Model (GBM) delivered the highest accuracy on new data, achieving F1-scores of 67% (low-risk), 45% (medium-risk), and 41% (high-risk) predictions.
Key risk drivers identified include precipitation and minimum temperature in March, and elevation. Specifically, risk is highest at altitudes below 1,000 meters when March rainfall is below 200 mm and soil temperatures range between 18°C and 27°C. During drier seasons, high-risk hotspots expand widely, while in wetter years risks are concentrated along coastal regions.
These results were recently published in the World Mycotoxin Journal.
Also read: ‘Levy on imported groundnuts now necessary’
A Practical Tool for Industry and Policymakers
Findings from the A-EWS are accessible through an interactive online dashboard, allowing users to visualize aflatoxin risk maps and make informed decisions.
The East Africa Grain Council (EAGC), a key partner and the regional voice for the grain sector, has endorsed the tool as a “useful platform for the grains industry.” The council highlights its importance in helping buyers identify safer maize sources and in providing location-specific recommendations for mitigation measures, including the use of Aflasafe® biocontrol products.
This targeted, data-driven approach replaces broad, inefficient “blanket” interventions with precise, evidence-based action.
“Using these maps, we can finally make aflatoxin visible,” said Jane Kamau of IITA. “The system equips policymakers, who may otherwise be unaware of local risks, with clear evidence to take action.”
Insights from the Developers
Scientists Reveal How AI Makes the ‘Invisible Enemy’ Visible
Lead scientist Dr. Francis Muthoni emphasised the tool’s transformative potential.
“Aflatoxin contamination has long been an invisible threat, difficult to monitor and manage at scale. Our AI-driven system changes that by providing clear, actionable risk maps that empower farmers, traders, and policymakers to act proactively rather than reactively,” said Muthoni.
“By integrating diverse data sources, satellite imagery, weather patterns, and historical contamination records, we’ve created a dynamic model that reflects the complexity of aflatoxin risk. This precision enables more effective targeting of interventions, saving crops, protecting health, and preserving market integrity,” he added.
Future Developments and Collaboration
Also read: CGIAR deploys tools to address plant pests and diseases
Expanding Beyond Maize: Future Plans for Comprehensive Crop Protection
The A-EWS project team notes that improving the accuracy of its predictions depends on expanding the volume and quality of contamination data. Upcoming steps include wider tool dissemination to gather user feedback, and the development of a back-end application that allows partners to upload new, standardized data to continuously refine the system.
Dr. Nancy Kirimi of the Kenya Agricultural and Livestock Research Organization (KALRO) is leading the creation of a standardized data collection protocol to ensure consistency and reliability across regions.
Plans are also underway to extend the system’s application to other vulnerable crops such as groundnuts and sorghum.
This initiative underscores a vital message for the upcoming summit: strong partnerships are essential to deploying market-ready innovations that can transform Africa’s food systems and enhance food safety for millions.
The Aflatoxin Risk Early Warning System stands as a transformative example of how cutting-edge technology and collaborative innovation are making the “invisible enemy” visible, empowering Africa to protect its staple crops, secure trade, and safeguard public health.
