Village-Scale Mapping of Agricultural Systems and Intensification Levels in Benin Using Geospatial Data and Machine Learning

Agricultural systems are rapidly transforming, making regular mapping critical for policy guidance. Field-based surveys remain costly and spatially limited, while satellite data represent a promising avenue for characterising agricultural systems. This study explores the use of freely available satellite and geospatial data to produce an exhaustive village-scale mapping of agricultural systems and their intensification levels in North and Centre Benin. An agricultural system typology, including intensification levels, was developed from a sample of 178 surveyed villages using hierarchical clustering of production system variables derived from field surveys. A Random Forest model was then applied to predict agricultural system types using 20 explanatory variables, including vegetation indices, landscape configuration metrics, environmental variables, and socioeconomic indicators. This approach enabled the spatially complete mapping of agricultural system types across the entire study area. Four agricultural system types were identified, reflecting a clear intensification gradient from low-input diversified agroforestry to mechanized and irrigated agropastoral systems. The Random Forest model achieved high accuracy (83.1%). MODIS-derived radiometric variables and environmental variables jointly dominated the prediction of agricultural system types, highlighting the key role of biophysical factors and farming practices in structuring landscapes. Except for distance to the nearest border, socioeconomic variables had a limited influence. This study provides a scalable and transferable framework for spatializing agricultural system diversity and intensification levels in heterogeneous landscapes. It reduces dependence on repeated field surveys and offers a robust foundation for policy-relevant, spatially explicit agricultural monitoring.