Highlights
•Predictive models for tree, soil and biodiversity attributes were studied.
•Tree crown cover was predicted most accurately using RapidEye and Landsat data.
•Prediction success might depend on correlation between the attribute and tree cover.
•The higher spatial resolution of RapidEye improved accuracy only slightly.
•Approach provides valuable information for land management intervention planning.

Forest management, agroforestry and tree planting are some of the key approaches to sustainable rural development, and climate change adaptation and mitigation in West African savannas. However, the planning of land management interventions is hindered by the lack of information at relevant spatial resolution. We examined predictive models for mapping various tree, soil and species diversity attributes with a comparison of RapidEye and Landsat imagery.
The field data was collected in the vicinity of the community-managed forest in southern Burkina Faso, where the main environmental threats are agricultural expansion and fuelwood extraction. The modelling was done using Random Forest algorithm. According to our results, tree crown cover and correlated attributes, such as basal area and tree species richness, were predicted most accurately. High spatial resolution RapidEye imagery provided the best accuracy but difference to medium resolution Landsat imagery was negligible for most attributes. Burn scar masked Landsat time series performed similar to dry season single date Landsat imagery, but the former avoids image selection and mosaicking, and could be less sensitive to artifacts due to the burn scars. The presented approach provides valuable information on important tree, soil and species diversity attributes for spatial planning of land management interventions.