Speaker
Description
Accurate estimation of aboveground carbon stocks in Mediterranean reforested forest stands is essential for supporting sustainable forest management, conserving vulnerable ecosystems, improving national greenhouse gas inventories, and contributing to the implementation of the Paris Agreement. However, conventional forest inventories are costly, labor-intensive, and spatially limited, making them insufficient for capturing the spatial variability of aboveground biomass (AGB) across large forest landscapes. This study presents a reproducible and integrated methodological framework that combines multi-source remote sensing data and machine learning techniques to estimate and map aboveground carbon stocks. The proposed approach integrates optical (Sentinel-2), radar, and satellite LiDAR observations with forest inventory plots and implements a stacking ensemble learning strategy based on several machine learning algorithms following a rigorous predictor selection procedure to improve estimation accuracy and model robustness. The originality of the framework lies in replacing the generic IPCC biomass-to-carbon conversion factor with a species-specific carbon fraction experimentally determined through laboratory analysis, thereby enhancing the reliability of carbon stock estimates. Furthermore, high-resolution carbon stock maps are produced together with an uncertainty assessment based on Monte Carlo simulations, providing a robust evaluation of estimation confidence. This standardized, transferable, and operational framework offers an effective solution for large-scale monitoring of forest carbon sinks and provides valuable support for carbon accounting, sustainable forest management, and climate change mitigation strategies in Mediterranean ecosystems.