Trust in a carbon credit depends on the rigor and integrity of the science underpinning it at the project level. For the Katingan Mentaya Project, this meant developing a robust measurement and monitoring system that integrates field ecology, satellite remote sensing, and machine learning. This needed to be done not only for the dynamics of a complex forest ecosystem, but also for a deep, millennia-old peat dome lying beneath it.
A tale of two carbon stores
Most forests store their carbon above the ground, in the trunks, branches, and leaves of living trees. The Katingan Mentaya Project does this, but it also sits on something far more important.
The Katingan Mentaya Project protects 157,875 hectares of forest in Central Kalimantan, Indonesia, and, in turn, the forest protects a deep layer of peat beneath. Peat is partially decomposed organic matter that has accumulated over thousands of years. In an intact, waterlogged state, this peat is inert. But when drained and exposed to air, as happens when a forest is cleared and the land converted, it oxidises and releases CO₂. Dry peat is also a dangerous fire hazard. Once lit, the fires can smolder underground for months, releasing even more CO₂.
Measuring and then protecting the carbon in both stores requires two distinct scientific approaches.
Wood through the trees: above-ground biomass
Calculating above-ground carbon begins with measuring individual trees. Our in-house Technical and Field Teams established 80 circular sampling plots, each with a 30 m radius, distributed using a stratified random sampling design. Within these plots, every tree with a trunk diameter of at least 10 cm at breast height (~130 cm above ground) is individually tagged, measured, and recorded, 17,788 trees in total. For a substantial subset, tree height was also measured using an ultrasonic hypsometer.
These measurements are then converted into biomass estimates using allometric equations, mathematical models that translate diameter and height into total biomass. The approach follows Chave et al. (2014) allometric model, one of the most widely used frameworks in tropical forest carbon science, with species-specific wood density values drawn from a global database. This process produced plot-level biomass estimates ranging from near zero in highly degraded areas to around 270 tonnes per hectare in the densest, least disturbed forest.
While plot samples provide a strong foundation for estimation, they cannot, on their own, capture the full spatial variability of a 150,000-hectare landscape. Forests are dynamic: degradation from illegal logging is uneven, and regrowth following disturbance varies across space. To generate a wall-to-wall, spatially explicit map of carbon across the entire project area, we integrate field data with satellite remote sensing.
From the ground to satellites
For a fuller, more accurate picture of the forest, we rely on a layered stack of Earth observation data, with each sensor contributing capabilities the others cannot provide on its own.
Global Ecosystem Dynamics Investigation (GEDI), mounted on the International Space Station, uses spaceborne LiDAR to fire laser pulses into the forest canopy and measure their return. This produces highly precise estimates of vegetation height and vertical structure—key inputs for biomass estimation.
Harmonised Landsat–Sentinel-2 (HLS) imagery combines data from Landsat program and Sentinel-2 mission to deliver high-resolution multispectral observations. These optical images capture forest condition, leaf area, and seasonal dynamics, providing rich contextual information about canopy health and variability.
Sentinel-1 mission, operated by the European Space Agency, is a Synthetic Aperture Radar (SAR) system that can collect data day or night, through cloud cover. This reliability is essential in tropical regions, where persistent clouds often limit the usefulness of optical imagery.
We also use PlanetScope PSB.SD imagery from Planet Labs’ SuperDove satellite constellation, which serves a complementary but distinct role. While GEDI, HLS, and Sentinel-1 feed into the biomass modelling pipeline, Planet data enables near-real-time monitoring of land cover dynamics. It allows us to detect forest disturbances, delineate precise boundaries of change, and monitor for illegal logging incursions within the project area.
With three-metre spatial resolution and near-daily revisit times, this dataset provides the granularity needed to identify small-scale disturbances that coarser-resolution satellites would miss. Once a land cover change event is detected and mapped using Planet data, biomass maps are then used to quantify the carbon impact of that specific loss.
It is worth noting that, like other optical systems (such as Landsat and Sentinel-2), Planet imagery is still affected by cloud cover, making SAR data an essential, weather-independent complement for continuous monitoring.
Alongside these publicly available data streams, we have integrated annual above-ground biomass maps produced by Chloris Geospatial. Chloris’s technology brings together satellite and airborne LiDAR data with machine-learning models to generate direct estimates of biomass stock and change at 30-metre resolution, with pixel-level uncertainty quantification, going back to the year 2000. These maps are built on global datasets and are peer-reviewed and independently validated – they represent the state of the art in satellite-derived biomass measurement. They are used in this project as a key input to the local modelling process, providing spatially comprehensive benchmark data that guides where field plots are most needed and what range of biomass conditions they should sample.
Ground truthing remains essential to calibrate remote sensing-based estimates by anchoring biomass model predictions to real field measurements, ensuring accuracy and reducing systematic bias. In other words, field plots provide high-accuracy ground truth but cannot cover the full landscape continuously. Satellite data covers the full landscape but requires local calibration to be reliable at the project scale. Therefore, the combination of the two, trained through machine learning, is far more accurate and more informative than either alone.
Training the model, machine learning and local calibration
Once all data streams have been processed, a two-stage machine learning pipeline is applied. In the first stage, GEDI-derived structural metrics, e.g. canopy height, cover, and vertical profile, are predicted across the full project area from the HLS and Sentinel-1 data. This step fills the spatial gaps left by GEDI’s sparse sampling track, producing continuous, wall-to-wall estimates of forest structure at 30-metre resolution across the whole project area.
In the second stage, the field plot measurements were used to train a localised model that predicted above-ground biomass from the full stack of structural and optical inputs. This local calibration step is critical: it grounds the global satellite data in the specific ecological conditions of the Katingan Mentaya Project’s peat swamp forest, even adjusting for species composition, local wood densities, and the particular relationship between canopy characteristics and biomass in this forest type. The resulting biomass map is substantially more accurate than a global model applied without local adjustment.
Because the satellite data is updated every year, the model can be rerun annually to produce an updated biomass map for the project area, tracking regrowth in recovering areas, detecting new instances of degradation, and monitoring the carbon consequences of any deforestation. Field plots are remeasured on a five-year cycle; between remeasurements, the remote sensing-driven annual update maintains the monitoring cadence. This means the carbon accounting is continuous, not periodic; it is always current, not estimated from a snapshot taken years ago.
Beneath the surface
Measuring the carbon stored in peat presents an entirely different challenge. Unlike a forest canopy, peat is not directly visible from space. Its depth varies dramatically—from a thin layer at the edges of the dome to several metres at its centre—and this depth determines both how much carbon it holds and how long it would take to be depleted under deforestation and drainage scenarios.
To address this, we developed a Digital Terrain Model (DTM) of the peat surface using elevation data from ICESat-2 mission (NASA) and the Global Ecosystem Dynamics Investigation (GEDI), combined with field-based peat depth measurements collected across the project area. Once accurately mapped, the characteristic dome shape of tropical peatlands allows peat thickness to be inferred at each location. By combining surface elevation with known drainage limits, depth can be estimated across the landscape.
Peat carbon emissions are driven primarily by water table depth. When peatlands are drained, the water table drops, oxidation accelerates, and CO₂ is released. The rate of emission depends on both the depth of the water table and the intensity of land use. Most projects—and even Intergovernmental Panel on Climate Change guidelines—rely on Tier 1 emission factors: broad averages based on land-use categories that do not capture site-specific variation.
For the Katingan Mentaya Project, we go further. Our technical team developed Tier 2 emission factors using a predictive model built from peer-reviewed data, incorporating both water table depth and land-use class as variables. (Methane and waterborne carbon emissions are also measured separately using dedicated emission factors.) This site-specific approach captures spatial variation in emission rates far more accurately than a single averaged value, and can be updated as new data—whether from the scientific literature or direct field measurements—becomes available.
The peat emissions model also accounts for subsidence, microbial decomposition, and peat combustion, simulating the gradual depletion of the peat carbon store year by year across the entire project area.
What this means for credit quality
The combination of these approaches, local-calibrated biomass mapping, annual satellite monitoring, deep-peat stratification, top-down continuous monitoring and Tier 2 soil emission factors, produces a carbon accounting framework that is more accurate, more conservative, and more transparent than the standards require.
Conservativeness is built in at multiple points. The wall-to-wall biomass mapping approach captures patchy degradation and natural disturbances that a field-plot-only approach would miss. The peat zone delineation applies a deliberate threshold: only areas of substantial peat depth are counted. The Tier 2 emission factors account for the site-specific rate of peat oxidation, rather than defaulting to the more favourable IPCC averages.
The annual update cycle matters for buyers and investors because it means the monitoring picture is current. A natural disturbance, for example, a flood, a fire, or an incursion, is detected in that year’s monitoring data and accounted for in that year’s credit issuance. During the 2021–2023 monitoring period, the project accounted for 461.8 hectares of deforestation, which is just 0.3% of the total project area, with the precise carbon impact of each event calculated and deducted from the credit total.
Looking forward
Every methodology has its limits, and science does not stand still. Permian Global and PT. Rimba Makmur Utama, the two organisations that manage the Katingan Mentaya Project, have recently completed the project’s first baseline renewal. This was an opportunity to incorporate advances in both remote sensing and peat science that were not available when the project was originally launched. New Tier 2 emission factors, updated biomass models, and a refreshed DTM all of these represent a deliberate effort to keep the science current.
For more news click HERE