Carbon stock estimation and REDD+ verification

Satellite lidar and SAR backscatter, fused carefully, can estimate above-ground biomass across millions of hectares and underpin credible REDD+ MRV. The methods work well in low-to-medium biomass forests; they hit hard physical limits in dense tropical canopy above roughly 100 Mg/ha.

Sensors

GEDI (Global Ecosystem Dynamics Investigation): Full-waveform lidar aboard the ISS. Three 1064 nm lasers produce footprints of approximately 25 m diameter, spaced 60 m along-track and 600 m across-track. Provides canopy height and vertical structure profiles used to calibrate wall-to-wall biomass models. Coverage is limited to latitudes between roughly 51.6° N and S, and the ISS orbit leaves spatial gaps that preclude direct wall-to-wall mapping without fusion with other datasets.
ICESat-2 ATLAS: Photon-counting lidar at 532 nm, six beams, 17 m footprint diameter, 0.7 m along-track photon spacing. Extends spaceborne canopy height sampling to higher latitudes (up to 88°) where GEDI does not reach, making it useful for boreal forest carbon accounting. Photon-counting mode requires more complex noise filtering than full-waveform approaches, and performance degrades in sparse canopies.
ALOS-2 PALSAR-2: L-band SAR (1.27 GHz) with scene resolutions from 3 m (spotlight) to 100 m (ScanSAR). L-band penetrates canopy more deeply than C-band, correlating better with above-ground biomass up to the saturation threshold, which in dense tropical forest typically falls between 80 and 150 Mg/ha depending on forest type and moisture conditions. Global mosaic products are produced annually by JAXA.
Sentinel-1 SAR: C-band SAR (5.405 GHz), dual-polarisation (VV+VH), 10 m resolution in Interferometric Wide Swath mode, 6-day revisit at the equator with both satellites. C-band backscatter saturates at lower biomass levels than L-band, often below 50 Mg/ha in closed-canopy tropical forest, limiting its standalone utility for high-biomass estimation. It contributes most usefully as a change-detection layer and as a predictor in ensemble models.

What REDD+ actually requires from a satellite

The UNFCCC REDD+ mechanism asks countries to demonstrate that forest carbon stocks are being maintained or increased relative to a reference level, and to report changes with enough rigour to satisfy international technical assessment. That demands three things: an estimate of carbon stock at a baseline date, a method to detect and quantify changes over time, and a documented uncertainty budget. Satellite data alone cannot supply all three, but it is increasingly central to all of them.

The standard approach converts above-ground biomass density (Mg dry matter per hectare) to carbon using a factor of approximately 0.47, then aggregates across mapped forest area. The critical input is therefore a spatially explicit biomass map, updated on a cycle that matches the country's reporting period, typically annual or biennial. Getting that map right in dense tropical forest is where the physics becomes genuinely difficult.

The saturation problem: why SAR backscatter has a ceiling

SAR backscatter increases with forest biomass because more woody material scatters more microwave energy back to the sensor. But the relationship is not linear and does not continue indefinitely. In closed-canopy tropical forest, the canopy and upper stems intercept and scatter most of the signal before it reaches the larger, lower stems that hold the majority of carbon. At that point, adding more biomass produces almost no additional backscatter change. The sensor is, in effect, blind to what matters most.

For C-band (Sentinel-1), this saturation typically occurs below 50 Mg/ha. L-band (ALOS-2 PALSAR-2) penetrates deeper and pushes the threshold to roughly 80 to 150 Mg/ha, depending on forest type, stand moisture, and surface roughness. In the Amazon or Congo Basin, where above-ground biomass can exceed 300 Mg/ha, both bands are saturated for a large fraction of the forest area. This is not a calibration failure or a data quality issue; it is a physical limit of microwave interaction with dense vegetation.

The practical consequence for REDD+ is that SAR backscatter alone cannot reliably distinguish a 200 Mg/ha stand from a 350 Mg/ha stand. Uncertainty in the high-biomass tail of the distribution is large, and that tail is precisely where the most carbon sits.

What GEDI and ICESat-2 add, and what they cannot do alone

Spaceborne lidar breaks the saturation problem by measuring canopy height directly from the time-of-flight of laser pulses. Canopy height correlates with biomass through allometric relationships derived from field inventory data. GEDI's 25 m footprint waveforms can resolve vertical canopy structure well enough to estimate Lorey's height (basal-area-weighted mean canopy height), which is among the strongest single predictors of above-ground biomass in most forest types.

The limitation is coverage. GEDI's 60 m along-track and 600 m across-track spacing means it samples perhaps 4 per cent of the Earth's land surface per year, and the ISS precession pattern leaves persistent spatial gaps. You cannot produce a wall-to-wall biomass map from GEDI alone. What you can do is use GEDI height retrievals as training and validation data for a model that predicts biomass across the full spatial extent using wall-to-wall predictors such as SAR backscatter, optical spectral indices, and terrain variables.

ICESat-2 ATLAS, with its six-beam photon-counting design and near-polar orbit, fills in the high-latitude gap where GEDI does not operate. Boreal forests in Canada, Russia, and Scandinavia hold substantial carbon stocks, and ICESat-2 provides the only current spaceborne lidar height data for those regions. Its photon-counting approach requires careful noise filtering but delivers dense along-track sampling that complements GEDI's waveform products in temperate and boreal zones.

Fusing the data streams into a biomass map

The fusion workflow that currently dominates published national-scale biomass mapping runs roughly as follows. GEDI height retrievals (and where relevant ICESat-2 heights) are matched to coincident SAR backscatter values from ALOS-2 PALSAR-2 annual mosaics and Sentinel-1 time-series statistics. A regression or machine-learning model, often a random forest or gradient boosting ensemble, is trained on those paired samples. The model is then applied to the wall-to-wall SAR and optical layers to produce a continuous biomass surface at 25 to 100 m resolution. Uncertainty is estimated by propagating model error, allometric equation uncertainty, and spatial sampling gaps through the prediction.

The approach works well in low-to-medium biomass forests, where the SAR predictors retain dynamic range. In high-biomass tropical forest, the model relies more heavily on ancillary predictors such as canopy cover from optical data, terrain, and climate variables, because the SAR signal has saturated. Reported uncertainties in those regions are correspondingly wider, often 30 to 50 per cent at the pixel level, though they can be reduced substantially when aggregated to national or sub-national reporting units.

Change detection for MRV purposes is somewhat more tractable than stock estimation. Deforestation and severe degradation produce large, detectable SAR and optical signals even where absolute biomass levels are uncertain. The carbon loss from a cleared hectare can be estimated by multiplying the pre-disturbance biomass estimate by the cleared area, accepting the uncertainty in the stock estimate but gaining confidence from the high accuracy of area measurement.

Honest limits and what field data still needs to do

No satellite-based MRV system currently meets the IPCC Tier 3 standard without substantial ground-truth input. Allometric equations, which convert field-measured stem diameter and height to biomass, vary by species, region, and forest type. Errors in those equations propagate directly into any satellite-derived biomass estimate. Countries with sparse national forest inventories will have wide uncertainty bounds regardless of how good the satellite data is.

Cloud cover is a persistent operational problem in tropical regions. Optical data for calibration and validation may be unavailable for months at a time in the wet season. SAR is cloud-transparent, which is one reason it anchors most tropical biomass mapping workflows, but its saturation limits mean it cannot carry the full analytical burden alone.

Satellize structures MRV analytics around these constraints, combining open-constellation data from GEDI, ALOS-2, and Sentinel-1 with client-supplied field inventory data to produce biomass maps with explicit, auditable uncertainty estimates. The approach mirrors the methods used in the Tonga crop-estimation programme: transparent inputs, documented assumptions, and outputs calibrated against in-country ground data rather than global defaults.

Any government or project developer preparing a REDD+ submission should budget for at least one cycle of field plot measurement to anchor the allometric calibration. Satellite data reduces the cost and increases the spatial coverage of MRV enormously, but it does not eliminate the need for boots on the ground.

Typical figures

GEDI footprint diameter	~25 m; along-track spacing 60 m, across-track spacing ~600 m
ICESat-2 ATLAS footprint	~17 m diameter; ~0.7 m along-track photon spacing; 6 beams
ALOS-2 PALSAR-2 resolution	3 m (spotlight) to 100 m (ScanSAR); annual global mosaic at 25 m
Sentinel-1 resolution	10 m (IW mode); 6-day revisit at equator (two-satellite constellation)
SAR frequency and saturation threshold	C-band (5.4 GHz): ~50 Mg/ha; L-band (1.27 GHz): ~80–150 Mg/ha (forest-type dependent)
Biomass map spatial resolution (fused product)	Typically 25–100 m depending on training data density and application
GEDI latitudinal coverage	51.6° N to 51.6° S (ISS orbit constraint)
ICESat-2 latitudinal coverage	Up to 88° N and S
Pixel-level biomass uncertainty (high-biomass tropical)	Typically 30–50%; reduces to 10–20% at national reporting-unit aggregation
Archive depth	GEDI: 2019–present; ALOS-2 mosaics: 2015–present; Sentinel-1: 2014–present

Analytics Satellize can run

Wall-to-wall above-ground biomass density map	GEDI/ICESat-2 height retrieval fused with ALOS-2 and Sentinel-1 backscatter via ensemble regression; allometric calibration against client field plots	GeoTIFF biomass surface (Mg/ha) with per-pixel uncertainty layer, at 25 or 100 m resolution, supplied as GIS layer
Carbon stock estimate by reporting unit	Spatial aggregation of biomass map over administrative or project boundaries; carbon conversion at 0.47 factor; uncertainty propagation per IPCC Tier 2/3 guidance	Tabular report with stock estimates, confidence intervals, and methodology annex suitable for UNFCCC submission
Biomass change and carbon loss quantification	Bi-temporal or multi-temporal biomass map differencing; area of disturbance from SAR and optical change detection; stock-times-area carbon loss calculation	Annual change report with spatial delineation of loss polygons and attributed carbon loss (tCO₂e), as GIS layer and tabular summary
SAR saturation diagnostic and uncertainty attribution	Backscatter-biomass response curve analysis by forest stratum; identification of saturation zones; sensitivity analysis across allometric equation variants	Technical memorandum quantifying saturation-related uncertainty by forest type and biomass class
GEDI sampling gap assessment	Spatial analysis of GEDI footprint coverage within project area; identification of under-sampled strata; recommendation for supplementary ICESat-2 or airborne lidar acquisition	Coverage map and sampling adequacy report
Reference level baseline biomass map	Historical ALOS/ALOS-2 mosaic time series combined with archived Sentinel-1 data; retrospective biomass modelling to UNFCCC reference period	Dated baseline biomass GeoTIFF with documented provenance for inclusion in Forest Reference Emission Level submission

Sources

Who does the work

We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.