AI model maps Senegal's smallholder crops with 84% accuracy using minimal data

A new University of Cambridge AI model, Tessera, identifies crops in Senegal with 84% accuracy, outperforming existing methods while using far less data and computing power, potentially aiding food security planning in the Global South. The model, tested in Senegal's groundnut basin, beat two standard satellite mapping methods and Google DeepMind's AlphaEarth, performing 28% better than the next-best model in one scenario. Its efficiency allows governments to extend sparse ground data across multiple years without new surveys, addressing a key barrier for local agencies. The research, published Sept. 29 in Environmental Research: Food Systems, gains urgency from a record El Niño threatening drought in West Africa. While accuracy dipped between 2018 and 2021, likely due to survey data quality, lead author Madeline Lisaius emphasizes the tool's accessibility as a step toward democratizing geospatial data.