Forecasting the future of forests with AI: From counting losses to predicting risk

Google Research · 8 months ago

Google Deepmind released ForestCast, a deep learning model that predicts future deforestation risk using satellite data instead of assembling region-specific geospatial inputs. The model uses vision transformers trained on Landsat and Sentinel 2 satellite imagery and achieved accuracy matching or exceeding previous approaches, with deforestation change history being the most informative input. The publicly available benchmark dataset and reproducible approach enables governments, companies, and communities to proactively target conservation resources to forests at highest risk before loss occurs.