Five tests to assess the viability and integrity of forest carbon digital MRV solutions
COP28 has galvanized renewed public and private sector commitment to forest conservation and restoration. Fit-for-purpose carbon data and monitoring systems can now play a vital role to ensure that these commitments are translated into impactful actions on the ground.
Today, many carbon standards and corporate GHG accounting protocols continue to rely on traditional monitoring approaches of land use emissions and removals, anchored in land use and land cover mapping approaches. Meanwhile, significant advances in Earth Observation science, AI and machine-learning technology have occurred. They enable cutting-edge algorithmic remote-sensing technologies that directly measure woody biomass dynamics remotely, with continuity in space and time. It allows to track forest carbon emissions and removals in a more comprehensive, robust, consistent and scalable manner. These technologies hold great promise to support governments, companies, civil society organizations, indigenous peoples and local communities with carbon surveillance systems for effective and transparent nature-based solutions.
Understanding whether a specific remote-sensing technology is fit-for-purpose is a challenging process though, especially for decision-makers who may not have a background in remote-sensing, forestry or carbon science. The selection process is even more challenging as there are currently no independent standards to certify the quality of biomass measurements. In the absence of such standards, decision-makers can consider the following five questions to assess whether a specific technology is fit-for-purpose.
1. How does the technology calculate and report uncertainty?
The Challenge: To monitor biomass changes, trees need to be “weighed”. But unless trees are cut and weighed, all biomass “measurements” are always estimates of the “truth”, no matter which approach is used. Uncertainty (error) metrics are therefore a key indicator to assess the quality of the estimate. However, uncertainty can be calculated and reported in different ways, making it challenging for end-users to assess the quality of the product. Some key questions to consider include: does the technology report location-specific uncertainty for both carbon stock and changes? At what confidence interval does the technology report uncertainty? And is the computed uncertainty comprehensive of all of the uncertainties present in the estimation process?
How Chloris does it: Chloris computes location-specific uncertainty for every pixel (10m and 30m resolution), reported as standard error. The standard error associated with the pixel-level change represents uncertainty in the change estimate at the 95% confidence interval level. The standard error is estimated from an error propagation analysis carried out at the pixel level across all layers in the time series. The propagation of error takes into account geolocation-, allometric-, and model-based errors. The standard error, pixel level estimates are aggregated to the site level in a process that takes into account the spatial autocorrelation and used to compute the overall site level uncertainty for the above-ground biomass stock and change. All uncertainty statistics are derived from sums of pixel values where the change (i.e., biomass gain or loss) was determined to be significant (p-value ≤ 0.05). The methodology is peer reviewed and published in: Baccini, et al. (2017).
2. Are the estimates validated against robust, independent data?
The Challenge: Because every biomass number is an estimate, the first question to ask when confronted with accuracy assessments is: compared to what? For machine-learning algorithms, the most commonly used approach reports model internal performance. It means that the model is tested against a subset of the data used to train the algorithm, but set aside and not used in the training process. Such model internal accuracy numbers are insufficient because they are not based on validation against data which is fully independent from that used to train the model. More insightful accuracy metrics are provided by methods that validate estimates against fully independent, higher quality data, not used to train the algorithm.
How Chloris does it: The Chloris product is validated against independent biomass measurements. Our White Paper (Feb 2023) includes the results of our first validation campaign, comparing Chloris against ALS data (Fig 1). In Q1 2024, we will publish the results of our second validation campaign, comparing our new models against independent ALS as well as against large sets of field data from academic networks. Recently, the ERS AGB Benchmarking Exercise has compared Chloris Geospatial data against ground LiDAR + UAV LiDAR measurements, as part of an industry-wide data benchmarking exercise. Chloris data was identified as the most consistent with higher quality biomass data.
3. Does the technology cover long time series?
The Challenge: Remote-sensing signals are inherently noisy due to issues such as ranging atmospheric interference or changes in plant phenology. To generate reliable remote-sensing estimates, reducing noise in the input data is essential, but not enough: noise also needs to be addressed in the output data. To do that and generate reliable estimates on biomass density and changes - the key to understanding emissions and removals - best practice approaches rely on long data time series. Rather than simply comparing two points in time, long time series create more reliable, more trustworthy trends.
How Chloris does it: Our pipeline produces a time series of annual carbon stocks and changes from 2000 to present at 30m resolution, and from 2017 to present at 10m resolution. This is done in two main steps: First, we create wall-to-wall maps of above-ground carbon stocks at annual time steps. Second, we process the annual estimates of carbon stock at each pixel using a Bayesian time series approach that removes noise in the time series and identifies statistically significant changes and trends in carbon stocks across time.
4. Does the technology deliver consistent and spatially explicit results?
The challenge: Delivering globally consistent biomass stock and change data is a necessary prerequisite to effectively compare the impact of interventions across multiple projects or jurisdictions. However, commonly used approaches based on land use cover/change estimates and emission factors typically suffer from inconsistent definitions of land use classes used across datasets, projects and countries.
How Chloris does it: Chloris uses the same methodology to directly estimate biomass stock and change for every 10m and 30m pixel, anywhere on the planet. Rather than relying on (man-made) definitions of forests and average emission factors, Chloris directly estimates the underlying ecological processes, i.e. biomass gains and losses and estimates forest cover/change over time based on locally-adapted biomass thresholds.
5. Is the technology scalable and cost-effective to use?
The challenge: The goal of MRV is to ensure and enable effective action on the ground. Remote-sensing solutions should be designed to be cost-effective and thereby contribute to optimizing budgets for actual forest conservation and restoration programs.
How Chloris does it: Our cloud-based infrastructure provides cost-effective data and analytics that are ready for customers to review in hours or days. We achieve cost-effectiveness and speed of production through three key elements: (i) our data processing infrastructure is highly efficient, (ii) our predictive models are trained at continental scale, and (iii) we use publicly available data as a starting point.
The Chloris approach in brief
Chloris Geospatial is a pioneer and leading provider of above-ground biomass stock and change and forest cover/change data. The Chloris software directly measures changes in biomass density of woody vegetation for any area of interest, anywhere on the planet. Chloris quantifies carbon emissions and removals for every pixel with 10m and 30m resolution, with annual data from 2000 onwards and quantified uncertainty for each pixel. The continuous mapping of spatially explicit biomass density changes in space and time enables more complete and robust quantification of carbon emissions (including those due to deforestation, degradation, as well as any other human or natural disturbances) and removals (including those achieved by intact, protected forests and restoration, reforestation, afforestation).
