Accurate information on where and how small farms are managed is essential for smart planning in agriculture. Yet, in many African regions, especially where farms are small and irregularly shaped, this kind of data is still hard to obtain. Without it, countries risk making decisions based on incomplete or misleading statistics, leading to wasted resources and ineffective agricultural support.

This work uses Artificial Intelligence and satellite imagery to automatically map smallholder farms in parts of the Great Rift Valley and Lake Kivu–Rusizi River Basin. By adapting pre-trained AI models with just a few local examples, we show it is possible to accurately map fields that are often missed by traditional image processing methods.

These AI-enhanced farm boundaries can support more targeted agricultural policies. They help governments and organizations understand how much land is being farmed, which crops are being grown, and where support is most needed. The results can guide more efficient input use, support climate-smart practices, improve yield tracking, and monitor greenhouse gas emissions at the farm level. This work provides tools that ministries of agriculture, environmental agencies, and partners can use to design more effective, evidence-based programs for small-scale farming.