Terrestrial, Airborne And Satellite LiDAR Enables Accurate and Scalable Carbon Content Measurements

Carbon credits are becoming a big business. Recently, Delta Airlines was sued by an unsatisfied customer because it claimed to be carbon neutral. The lawsuit asserts that this was not, in fact, the case. The complaint alleges that the airline relied on carbon offsets that were not real. Companies buy carbon credits in the commodities market to neutralize carbon releases with projects designed to absorb carbon dioxide out of the air. Claiming carbon neutrality is important – consumers feel better about their choices and are willing to pay more for products and services that claim neutrality. The issue is about trust – and an unbiased entity that can accurately rate these projects in terms of true carbon absorption capability – and track it in time.

Conventional methods (Figure 1) rely on tree diameter and height measurements to estimate stem volume or tree carbon sequestered and stored in the vegetative tissues of the growing trees. In simple terms, that means using a tape measure to measure the diameter of trees in a sample plot as input to an allometric function, which outputs an estimate of the aboveground carbon absorption capacity or biomass of a project area (for example, the Amazon forest). The United Nations, for example, uses this approach as they push member nations towards environmental stewardship and carbon neutrality. These methods are intensive in terms of manual effort, prone to error and sparse in terms of data gathered.

Remote sensing using LiDAR (Light Detection and Ranging), cameras and radar are also being deployed in terrestrial (static and mobile), drone and airborne platforms to create color maps and 3D maps of biomass projects. Satellite imagery is also used. This data can then be used to estimate the carbon absorption capacity of different biomass projects. Ground-level data provides higher accuracy and resolution but is expensive to collect over large areas. Higher altitude sensing yields more efficiencies in terms of project area covered/hour but has lower resolution and accuracy. Satellite data is even lower in resolution but continuously monitors large areas as they orbit the Earth. The challenge is bridging these different altitude data sets to achieve the right balance of accuracy, scalability and freshness.

Enter Sylvera (Sylva is Latin for “forest”; era is about time – “forests in time”), a London, U.K.-based startup launched in 2020 whose mission is “to incentivize investment in real climate impact” with better data about carbon storage in nature (Figure 2). It disrupts traditional approaches using MSL (Multi-Scale LiDAR), ML (Machine Learning) and satellite data to achieve the optimal balance of accuracy, resolution, scalability and data freshness. In addition, Sylvera builds software that independently and accurately automates the evaluation of carbon projects that capture, remove, or avoid emissions. The field carbon data calibrates the company’s machine learning models to evaluate and deliver independent global carbon ratings to different projects that capture, remove, or avoid emissions. The ratings can be used to price the value of carbon projects in commodity markets (similar to S&P or Moody’s ratings on the quality of bonds).

Sylvera has raised $38M to date and just announced a Series B investment of $57M with Balderton Capital leading the round with participation from existing investors Index Ventures, Insight Partners, Salesforce Ventures, Speedinvest, Seedcamp and LocalGlobe, and new investors Fidelity Strategic Ventures, Bain & Company and 9Yards Capital. It currently has ~150 employees, has forged partnerships with S&P Global, and acquired clients ranging from major financial services institutions to sovereign governments.

Sylvera’s approach is summarized below:

1) Identify regions of interest (typically ~ 50,000 hectares) representative of forests regionally or nationally, ensuring these areas capture the variations in forest structure, state, succession and taxonomy over which MSL data are acquired. The selected area captures as much of the variation in AGB (above-ground biomass) as that encountered across the wider area (e.g., biome). This includes, for example, data collection across non-forest (e.g., agricultural land etc.), through to old-growth forests.

2) TLS (Terrestrial Laser Scanning) is used to create high-resolution 3D point clouds (5-10 m resolution) in multiple one-hectare (2.5 acres or 10,000 m²) areas within the selected area in the biome. The point clouds enable the estimation of individual tree volume, above-ground biomass, and carbon. The accuracy of these TLS-based methods is validated using direct, destructive weighing measurements of a select number of trees. Sylvera claims potential accuracy of as close as 3% relative to these reference harvest data generated by destructive testing. In contrast, conventional allometrics (that relate tree size to mass via measurements such as stem diameter) have close to a 40% error, especially for larger trees. This is because conventional techniques are highly manual, sparser, and error-prone. Syvlera’s approach automates the process, creating richer, dense and higher-resolution data.

3) UAV-LS (Unoccupied Aerial Vehicle Laser Scanning (UAV-LS) is then conducted over coincident and larger areas (e.g., thousands of hectares). ALS (Airborne Laser Scanning) uses drones or conventional aircraft at higher altitudes over larger land masses to acquire LiDAR data. The aerial data are obtained using a commercial 1550 nm, polygon scanned, ToF (Time-of-Flight) LiDAR system designed especially for aerial 3D mapping. Various metrics describing forest structure can then be retrieved from the resulting point clouds (which have hundreds of points/m²), such as tree cover, canopy height maps and 3D distributions of plant material.

4) Machine learning (ML) is then used to predict forest above-ground biomass across the UAV-LS and ALS sections from the above metrics. The TLS-derived estimates from Step 2 provide the training data. ML correlates ABG (above-ground biomass) with changes in forest structure. Rigorous cross-validation methods are used to understand and quantify the uncertainties in these approaches.

5) The final step to generating region-scale estimates of forest ABG and carbon stocks is to use multimodal satellite data (e.g., radar, lidar and optical), whereby the MSL data train new ML models. The satellite data can be sourced from commercial and government satellite operators. ML models can monitor entire project areas (see Figure 3) over time. Using satellite imagery provides excellent scalability and data freshness, resulting in monitoring and regular updates of the carbon rating of specific projects.

The current focus is collecting data from all the different forest types (e.g., tropical, subtropical, temperate and boreal, and their various subcategories) and relying on ML models and satellite data to monitor AGB and carbon content. Changes in biome vegetation (different from that used in the earlier ML model development) may warrant repeating the above steps, although this is likely over a longer time frame (5-10 years).

Based on the MLS, satellite data, and ML, Sylvera provides customers with carbon ratings of the type shown in Figure 4.

The commodities market can use such ratings to price carbon credits for projects similar to S&P, Moody’s bonds or mutual funds ratings. It also gives buyers and regulating bodies confidence and assurance to assess their carbon offset requirements and certifications.

Allister Furey, the founder and CEO of Sylvera: “Making progress against societal net zero starts with having the data to measure actions and their impact on the climate. We cannot understand the real impact of projects and investments without understanding how much carbon is in the world. That’s why Sylvera has developed the most accurate methodologies of in-forest biomass measurement.”