A tree level carbon inventory in Cyprus enabled by the fusion of ground data and high-resolution aerial imagery with a deep learning model

Accurate quantification and long-term monitoring of tree biomass and carbon stock at high spatial resolution is critical for understanding forest ecosystems dynamics and supporting land-based mitigation efforts. This study presents the first high spatial-resolution assessment of tree biomass and carbon stock within and outside forests across the Republic of Cyprus, in the Eastern Mediterranean. This is achieved through the integration of field inventory data and 50 cm airborne optical remote sensing images with a deep learning model. The model was trained and locally adapted to detect, count, and map the crown area of individual trees, enabling tree-level mapping for the entire island. To estimate tree biomass, new allometric equations specific to terrain aspects were developed using field data and a high-resolution digital terrain model, compatible with remote sensing–derived tree crown area and height. These equations allowed the computation of individual tree biomass and carbon stock from their crown area, as well as aggregated tree densities and per-hectare values across forest and non-forest land-use types. This new database allows quantifying the number of burned trees, burned tree cover, and associated biomass and carbon stock losses—going beyond traditional burned-area assessments. To estimate biomass loss from fire, satellite-derived burned areas maps since 2016 are combined with our tree-level biomass maps. This study establishes a scalable, data-driven framework for precision long-term forest monitoring at country level. It provides the first detailed quantification of Cyprus’ carbon resources at tree level, and sets the foundation for improved national greenhouse gas reporting, post-fire restoration planning, and cost-effective forest management.