AI for Empowering Smallholder Farmers in Africa
Degas financed over 64,000 farmers since its inception and doubled their incomes. As we scaled to support tens of thousands of farmers across Ghana, we faced a challenge; how do we monitor and support thousands of fields at once?
The answer was space.
We turned to satellite imagery, the only scalable and cost-effective way to track farms remotely. To make this vision possible, we built a world-class team of machine learning researchers and satellite data engineers. Together, we developed some of the most advanced geospatial AI models in the world.
Our first breakthrough came with Degas 100M, a foundation model trained on millions of satellite images from the Sentinel-2 constellation. Benchmarking showed that our model outperformed NASA/IBM’s Prithvi across all major tasks in the PhilEO benchmark, including land cover classification and segmentation. See our paper.
But agriculture is not static. Farms evolve over time, and good decisions require a memory of the past. That’s why we built the first large-scale model designed to process five years of historical satellite data. Trained on over half a million long time-series, our model adapts seamlessly to a range of agricultural tasks—from predicting yield to mapping crop cycles—using deep temporal context.
Our work doesn’t stop with optical imagery. In partnership with Japan’s Remote Sensing Technology Center (RESTEC), we built a radar-based foundation model using ALOS data. This system achieved over 80% accuracy in flood detection—a vital tool for disaster response.
We’ve also demonstrated how AI can directly impact food security. In Thailand, we deployed our geospatial models to detect cassava mosaic disease, which causes hundreds of millions in annual crop losses. Our model achieved over 90% accuracy in field trials and was extended to map soil pH and organic matter content across the country, all with minimal ground truth input.
With our foundation models, we’re creating a new kind of agricultural intelligence—scalable, adaptable, and deeply rooted in real-world operations. Degas is where boots-on-the-ground logistics meet state-of-the-art machine learning.