Forest carbon estimates used to underpin multi-billion-dollar carbon markets are wrong by up to 50% for large trees, which hold most of the world's forest carbon. Current methods rely on measuring tree diameter and height in the field, then plugging those numbers into simple equations that estimate wood volume. This process systematically underestimates carbon in big trees, skewing global carbon accounts. The problem matters because carbon markets—which fund reforestation and forest protection projects—need accurate data to verify that a tonne of carbon claimed is actually a tonne stored. Without it, buyers cannot trust net-zero claims, and the market cannot scale to meet the Paris Agreement targets. This £1.5 million project, led by Sylvera with University College London and NASA’s Jet Propulsion Lab, will build a new training dataset for machine learning models that estimate above-ground biomass directly from satellite imagery. The team will capture ground-truth measurements accurate enough to calibrate satellite data, replacing the current inventory-derived estimates. If successful, the approach could unlock reliable, global-scale carbon accounting, allowing carbon markets to support billions of pounds in forest restoration and planting.
View original technical description
The world's current understanding of how much carbon is stored in the Earth's forests has been shown to be **fundamentally inaccurate**, with **serious implications for the fight against climate change.** These issues have real world impact. The world needs Nature Based Solutions projects (i.e. reforestation, forestry protection etc.) **to scale more quickly**. The Paris Agreement set a 2 degree warming scenario. We are currently on track for a 3 degree warming scenario, with all the catastrophic consequences this entails. All the various players in the Carbon Market (project developers, brokers/intermediaries and sellers) **need accurate, validated data to ensure frictionless, increased trade, proving their net zero claims.** However, **current** measurement techniques rely on **outdated and biased carbon estimations** which approximate biomass, and hence carbon, from tree diameter and height and inaccurate sampling. Machine Learning (ML) models offer the capability to accurately estimate Above Ground Biomass (AGB) from the current and imminently available raw Satellite Earth Observation (EO) data at a global scale. **However, they need accurate, well-calibrated training data** with which to train ML models with, **which is currently absent and is the focus of the project.** The **currently** available data to train models that infer AGB from satellite EO data is "inventory derived AGB data". This is gathered by manually measuring two standard parameters: tree diameter and tree height, and then estimating the tree volume and hence biomass using an allometric model which relates those tree measurements to volume, with a simple linear model. **This process does not accurately quantify biomass** and has recently been demonstrated to exhibit systematic bias (**up to 50%**) for quantifying carbon in large trees \[7\], which dominate the stores of carbon in forests. Likewise, very recent approaches using Space or Airborne LIDAR and the forthcoming ESA BIOMASS Synthetic Aperture Radar (SAR) mission, offer potential for excellent AGB inference accuracy \[6\], but are also constrained by the **low accuracy of ground measurements**. This is an exciting £1.5 million project led by Sylvera in conjunction with its partners **University College London and the NASA Jet Propulsion Lab** to push forward the state-of-the-art in Earth Observation technology. The team will capture accurate data on the carbon stored in the world's forests, with the aim of revolutionising global carbon markets, allowing them to **scale and support billions of dollars of forest restoration and planting.**
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know