Using Generative AI and Foundation Models to Predict Above Ground Biomass
summary
Claims a scalable method to estimate above ground biomass density worldwide by fine‑tuning a geospatial foundation model on multispectral satellite imagery and LiDAR‑derived labels; uses HLS harmonized Landsat Sentinel‑2 30 m tiles plus Pulsar radar and NASA GEDI LiDAR height and AGB labels ingested and processed with the geospatial Foundation Model and pyge stack. Produces global 30 m maps with uncertainty, outperforms Random Forest and U‑Net baselines, but underpredicts very high AGB and shows bias toward central distribution; practical for NetZero accounting.
AI-written summary · about
chapters (11)
- 0:00AI for carbon sequestration
- 3:13Biomass prediction goals
- 6:25Satellite imagery approach
- 10:37Biomass estimation challenges
- 18:11Cloud handling and seasons
- 27:03LiDAR challenges
- 29:56Data harmonization
- 31:03Model experiments
- 32:31Performance metrics
- 38:27Global biomass maps
- 41:35Summary and Q&A
related talks
48mTaking the planet’s pulse: AI to empower adaptation to global change
27mSmart Forestry: AI in the Service of Nature
41mAbby Seneor | AI Track | Compass AI & Data Summit 2025
28mGeoAI: Evolution of AI-Based Tree/Vegetation Detection Models Using Imagery
32mMeasuring Vegetation Health to Predict Natural Hazards
09mGen AI Workflows for Climate Risk Modeling in Food Systems
36mWatching Millions of Trees with Python: For Impact and Profit
48mSetting the stage for AI for biodiversity