Global Ecosystem Dynamics Investigation-enabled machine learning for aboveground blue-carbon mapping in the Cua Dai estuary, central Vietnam

Accurate blue-carbon assessment in fragmented estuarine landscapes remains challenging because multispectral optical data may saturate over dense canopies and plot observations are often sparse. This study developed a Google Earth Engine workflow integrating Global Ecosystem Dynamics Investigation (GEDI) spaceborne Light Detection and Ranging (LiDAR), Sentinel-1 C-band synthetic aperture radar (SAR), Sentinel-2 multispectral imagery, Shuttle Radar Topography Mission (SRTM) topography, principal component analysis (PCA), and machine-learning models to estimate aboveground blue-carbon stocks in the Cua Dai estuary, central Vietnam. After strict quality, sensitivity, and land-water screening, 2,781 GEDI top-of-canopy height (TCH) footprints and 6,711 GEDI aboveground biomass density (AGBD) footprints were retained as spaceborne reference observations. Random Forest (RF), Gradient Boosted Trees (GBT), Support Vector Machine (SVM), and Classification and Regression Trees (CART) were compared. RF produced the most stable performance, with root mean square error (RMSE) values of 17.32 Mg ha -1 for AGBD and 5.45 m for TCH. The resulting map estimated a mean aboveground carbon (AGC) stock of 13.659 Mg C ha -1 and a total AGC pool of 109,472.12 Mg C across the vegetated landscape. High-carbon hotspots were concentrated in mangrove and riparian alluvial zones in the southwestern sector. Because no local plot-based biomass inventory was available, the outputs should be interpreted as GEDI-calibrated, aboveground carbon estimates rather than fully field-validated total blue-carbon stocks. The workflow provides a rapid and scalable basis for restoration targeting, ecosystem-service valuation, and future measurement, reporting, and verification systems.