Madeline Lisaius was staring at a satellite image of Senegal’s groundnut basin, where tiny farms stretch across the land like patchwork, when she realized something remarkable: a new AI model had correctly identified millet, groundnut, and cowpea fields with 84% accuracy—using far less data than anyone thought possible.
In a country where most food comes from small, rain-fed farms vulnerable to drought and climate swings, knowing what crops are growing where can mean the difference between hunger and security. Yet for years, reliable crop data has been out of reach, gathered only through slow, costly field surveys every few years. Now, Tessera, an open-source AI developed at the University of Cambridge, is changing that.
Trained on satellite images, Tessera maps crops across Senegal by turning each 10-meter patch of land into a digital 'fingerprint' that captures how it changes over time. With just a fraction of the computing power and labeled data, it outperformed existing methods, getting it right 84% of the time. In one test, it was 28% more accurate than the next-best model. Unlike other systems, it also works well when applied to different years—meaning governments can update crop maps without starting from scratch.
"Accurate and up-to-date crop statistics can guide food security planning and help decide where best to target support," said Lisaius, who led the research as a doctoral student. That’s especially critical now, as Senegal faces one of the strongest El Niño events on record, threatening rainfall patterns and harvests.
The United Nations World Food Programme is already exploring how to use tools like Tessera to monitor food risks faster and more precisely. For smallholder farmers who grow food on plots no larger than a football field, this technology could mean earlier warnings, smarter aid, and better chances of surviving a bad season.
The study, published in Environmental Research: Food Systems, shows that with AI like Tessera, countries in the Global South don’t have to wait for expensive infrastructure to get the data they need. The model is open-source, so governments and NGOs can use it now to create their own crop maps.
As climate threats grow, so does the value of knowing what’s in the fields. With models like Tessera, a clearer picture of food security is finally coming into view—one pixel at a time.