Above Ground Biomass

Accurately estimating above-ground biomass is critical for carbon certification.

The project integrates NASA’s GEDI (Global Ecosystem Dynamics Investigation) LiDAR data into a machine learning biomass model. GEDI is a spaceborne laser instrument that provides detailed 3D measurements of forest structure (canopy heights and vertical profiles) (GEDI Lidar | NASA Earthdata). From these measurements, the GEDI science team derives above-ground biomass density (AGBD) at sample footprints on the ground (Estimate biomass using GEDI and Landsat data | Learn ArcGIS). The startup uses these GEDI-derived biomass points, along with satellite imagery and environmental data, to train a predictive model (using algorithms like random forest regression or a deep neural network). Input features include multispectral indices (e.g., NDVI from Sentinel-2 or Landsat), canopy height metrics, and topography, which the model learns to correlate with biomass. The trained model can then estimate biomass for the entire project area, not just GEDI sample spots. This AI-driven approach dramatically speeds up biomass mapping compared to manual forest inventories. It also improves accuracy – recent research with similar methods showed an attention-based deep learning model achieved an R² of ~0.66 and lower error in biomass prediction, outperforming traditional models ([2311.03067] Forest aboveground biomass estimation using GEDI and earth observation data through attention-based deep learning). In our context, we expect the ML biomass estimates to be within ~10-15% of field measurements, which is sufficient for carbon credit calculations. The biomass outputs are continuously refined by comparing against new GEDI passes and occasional ground truth plots, ensuring robust estimates of carbon sequestration over the years.