Accurately measuring the amount of biomass (living plant material) in forests is important for tracking carbon storage and supporting climate-friendly forest management. However, the forests along the southeastern U.S. Coastal Plain, such as those in southeastern North Carolina, have not been well mapped in terms of above-ground biomass (AGB).

To address this, we established machine learning models that combine different types of satellite and aerial data to estimate forest biomass over our study area. The data used includes images from multiple satellites (Sentinel-1 and Sentinel-2), topographic maps, and tree height information from low-cost high resolution stereo aerial photography.

We tested different combinations of data and found that adding tree heights from aerial stereo photos made the model significantly more accurate. Our study also used an explanation tool (called SHAP) to understand which data features were most important in making accurate predictions. The final map, which depicts forest biomass at a 30-meter resolution, can support carbon monitoring, inform sustainable forestry practices, and guide related policy decisions.