Quantifying carbon stocks and tree community composition in tropical forests through integrated satellite and UAV analysesAuthor

Monitoring tropical ecosystem services such as carbon stocks and biodiversity with remote sensing is essential for addressing climate change and biodiversity loss, but collecting ground truth data is costly. We investigated whether Unmanned Aerial Vehicles (UAVs) can reduce these costs. First, we developed a method to estimate Above-Ground Carbon (AGC) and a biodiversity indicator (mixing ratio of pioneer and late-successional species) from UAV-RGB images. Over 500 ha of imagery were captured in lowland tropical forests across four Forest Management Units (FMUs) in Sabah, Malaysia. Using canopy height and late-successional dipterocarp abundance, we built regression models (R² = 0.80, 0.38 for AGC and biodiversity) and extrapolated them across the imagery. Second, we tested whether adding UAV-based ground truth improves satellite-based models. We built machine learning models using Landsat metrics and tree inventory data outside the FMUs (n = 287). Accuracy was low without local data (R² = 0.43 and 0.46 for AGC and biodiversity). Adding UAV-based data from the FMUs (n = 934) increased accuracy (R² = 0.51 and 0.48), comparable to using local tree inventory data (n = 107; R² = 0.53 and 0.60). Combining UAV and satellite data enables effective monitoring of ecosystem services while reducing ground truthing costs.