AI Maps Senegal Smallholder Crops With 84% Accuracy Using Limited Data

Researchers at the University of Cambridge have demonstrated that their open-source artificial intelligence model, Tessera, can accurately map smallholder crops in Senegal using minimal training data. Published on September 29 in Environmental Research: Food Systems, the study highlights how accessible satellite analysis can support food security monitoring in the Global South. Agriculture in Senegal relies heavily on rain-dependent smallholder farms, leaving millions vulnerable to climate volatility. Traditional crop monitoring depends on costly ground surveys or computationally intensive satellite analyses, methods that are often impractical for local governments and humanitarian agencies. To address this gap, the Cambridge team applied Tessera to the groundnut basin, a critical agricultural zone. The model processes a full year of satellite imagery to convert each ten-meter land segment into temporal data embeddings. A lightweight algorithm then requires only sparse calibration points to generate large-scale crop classifications. In testing across multiple years, Tessera achieved an 84 percent accuracy rate, outperforming comparable mapping techniques by up to 28 percent while consuming significantly fewer computational resources. Crucially, the model maintained high reliability when trained on single-year datasets and applied to subsequent years, eliminating the need for annual ground surveys. Lead author Madeline Lisaius noted that the technology enables continuous monitoring between sparse data collection cycles, providing more precise crop statistics than baseline methods. The findings carry immediate relevance as West Africa navigates one of the most intense El Niño events on record, which threatens to disrupt regional rainfall and exacerbate drought conditions. The World Food Programme has indicated that geospatial data and artificial intelligence could strengthen food security monitoring and accelerate decision-making for vulnerable communities. By lowering technical and financial barriers, Tessera allows NGOs, government bodies, and research institutions to generate independent crop maps without reliance on proprietary systems. While the study confirms the model effectiveness, researchers acknowledged limitations, including a modest accuracy decline between 2018 and 2021 that correlated with ground survey data quality. The model also did not evaluate fields hosting multiple overlapping crops, a factor that could influence precision in highly diversified agricultural zones. Despite these constraints, the primary advancement lies in deployment accessibility. Lisaius emphasized that Tessera does not need to be flawless to be transformative, as it democratizes advanced geospatial analysis for regions previously excluded from precision agriculture technologies. The approach builds on earlier successful deployments in European farmlands, now scaled to support resilient food systems in developing economies.