New Model Estimates Forest Biomass Using Multi-Sensor Data.

Researchers have developed an operational framework for quantifying forest above-ground biomass (AGB) using a single, globally trained convolutional neural network (CNN). This model integrates data from multiple sensors, including optical (Sentinel-2), C-band SAR (Sentinel-1), L-band SAR (ALOS-2 PALSAR-2), and terrain (DEM) information. The CNN is initially trained on GEDI Level-4A biomass reference data from diverse regions and seasons, allowing it to learn persistent woody structures rather than single-date appearances. A key innovation is the lightweight empirical field-calibration workflow, which adapts the global model to new landscapes without requiring full retraining. This involves using a small number of local field plots to apply a scale-and-bias correction, aligning the global predictions with ground truth in specific regions. This approach significantly improves local validation performance, outperforming both the uncalibrated global model and existing products like ESA CCI Biomass, making carbon accounting more credible and mitigation strategies more actionable.