Mapping the spatial extent of agroforestry lands across Africa by characterizing tree spatial arrangements

Agroforestry is a land-use practice that integrates trees into agricultural lands, supporting food security and rural livelihoods while delivering both climate adaptation and mitigation benefits. However, current large-scale assessments often approximate the spatial extent of agroforestry using coarse tree canopy cover or biomass products that cannot characterize tree spatial arrangements within croplands, potentially obscuring structurally distinct agroforestry systems. Here, we produced a 100 m resolution map of the spatial extent of agroforestry lands in Africa for 2019, using 10 m resolution tree canopy cover data derived from PlanetScope, together with Sentinel-2 imagery. This map explicitly delineates croplands where agroforestry systems are practised by distinguishing tree spatial patterns within croplands, focusing on irregularly scattered trees and trees aligned in rows. Scattered trees were identified using tree-pixel distance metrics, whereas trees planted in rows were detected using a U-Net model trained with ∼38,000 manually labelled tree-row features. The identified agroforestry lands with scattered trees and trees in rows on cropland were validated against ground-based sites collected from the literature and Google Earth imagery, yielding an overall accuracy of 77.2%. This agroforestry dataset with explicit spatial tree arrangements can serve as a baseline for assessing the environmental impacts and changes in crop yield associated with agroforestry ecosystems, as well as for supporting the development of process-based agroforestry models.