Straightforward model-based approach using only field data and open-source maps to improve carbon stock estimates for REDD + projects.

Accurate estimation of forest biomass at property-level finds diverse applications, being particularly important for REDD + projects (Reducing Emissions from Deforestation and forest Degradation). This study presents a straightforward method for improving carbon stock estimates by integrating geolocated field plot data with open-source large-scale maps. We evaluated the performance of simple predictive models using spatial coordinates and global maps as covariates. Our results demonstrate that even without advanced remote sensing data and complex modeling techniques, incorporating spatial information and open-source data can substantially improve carbon stock estimates. Spatial coordinates and global map information significantly enhanced predictions with a 31.9% decrease in MAE for areas near field plots (i.e., predictions inside the REDD + project), and 18.6% decrease in greater distances (i.e., predictions for the broader region outside the REDD + project). Moreover, our approach allows for the generation of high-resolution wall-to-wall carbon stock maps for entire REDD + project areas, even in the absence of high-quality local remote sensing data. We conclude that large-scale maps, when properly calibrated with local field data, are invaluable for improving carbon stock predictions in tropical forests. Our method is widely applicable, providing a practical solution for users interested in enhancing their carbon stock estimates at a local scale.