An African Tree Crown–Stem Allometry Dataset for Monitoring Trees Outside Forests and Landscape Restoration

Abstract. Very high-resolution (VHR) satellite imagery and advances in machine learning increasingly enable individual tree crowns to be mapped across large landscapes, creating new opportunities for census-based estimation of tree biomass and carbon. A fundamental limitation, however, is that satellite imagery can measure crown dimensions but cannot directly observe diameter at breast height (DBH), the principal predictor used in conventional tree biomass allometry. This measurement gap is particularly important in African landscapes dominated by trees outside forests (TOF), where trees are spatially dispersed, conventional forest inventory sampling is difficult to scale, and ground datasets containing concurrent measurements of crown and stem dimensions remain scarce. Here, we present an African individual-tree crown–stem allometry dataset designed to provide the empirical link between remotely observable crown dimensions and the stem measurements required by established biomass allometric models. The dataset was collected using a standardized field protocol across landscapes associated with African Forest Landscape Restoration Initiative (AFR100) projects in Kenya, Ghana, Rwanda, and Malawi. The sampled landscapes encompass agroforestry, reforestation, assisted natural regeneration, and other restoration and land-management systems representative of TOF conditions. Following quality control, the dataset contains 3,493 individual trees with concurrent measurements of DBH, two perpendicular crown diameters, tree height, species identity, geographic location, and associated plot and landscape identifiers. Field measurements were designed for spatial linkage with individual tree crowns delineated from VHR satellite imagery, providing calibration observations for translating remotely sensed crown projected area (CA) into stem diameter.

We demonstrate this application by fitting power-law DBH–CA transfer functions and evaluating their predictive performance across sites and countries. All site-specific relationships were significant (p < 0.001), with R² values ranging from 0.44 to 0.85 and generally small prediction bias. Variation in model coefficients and predictive error among landscapes demonstrates substantial ecological and structural heterogeneity in crown–stem relationships and emphasizes the value of geographically distributed calibration data rather than reliance on a single universal allometric relationship. A hierarchical synthesis of 19 site-specific models further demonstrates how these data can be used to derive a generalized crown-to-stem relationship while retaining information on among-landscape variation. The principal contribution of this dataset is therefore to provide a measurement bridge between individual-tree remote sensing and conventional stem-based biomass allometry. VHR imagery and machine-learning methods can delineate individual crowns and measure their projected areas; crown-to-stem transfer functions calibrated with these field data can then estimate DBH; and established DBH-based allometric equations can subsequently be used to estimate biomass and carbon for each remotely detected tree. This framework extends individual-tree allometric scaling from field plots to landscape-scale tree censuses while maintaining compatibility with established forest mensuration and biomass estimation methods. The dataset has applications beyond the models demonstrated here, including development and validation of alternative crown–stem and crown–biomass models, analysis of tree structural and species-level variation, ecological scaling studies, and calibration of individual-tree remote-sensing products. It provides a reusable observational resource for advancing the inclusion of TOF in Forest and Landscape Restoration monitoring, carbon accounting, and measurement, reporting, and verification across African landscapes.