Spaceborne lidar calibration of above-ground biomass maps

Wall-to-wall biomass maps from SAR and optical sensors carry systematic bias that only direct height measurements can correct. GEDI and ICESat-2 provide those measurements from orbit, with real constraints on where and how densely.

Sensors

GEDI (Global Ecosystem Dynamics Investigation): Full-waveform lidar mounted on the ISS. Three lasers produce eight ground tracks with 25 m footprint diameter, spaced roughly 600 m apart across-track and 60 m along-track. Delivers canopy height, relative height metrics (RH50, RH98) and plant area index. Latitude ceiling: 51.6° N/S. Coverage is sparse and non-contiguous by design; it samples, it does not map.
ICESat-2 ATLAS: Photon-counting lidar in a 92° inclination orbit, giving near-global coverage including boreal and high-latitude forests excluded from GEDI. Six 11 m beams in three pairs. ATL08 land and vegetation product provides canopy height at 100 m segments. Denser along-track sampling than GEDI but similarly non-contiguous across-track.
Sentinel-1 SAR (C-band): 10 m resolution, 6-day repeat at mid-latitudes in IW mode. C-band backscatter saturates at relatively low biomass levels (roughly 50–100 Mg ha⁻¹ in dense tropical forest), limiting its standalone utility for high-biomass stands. Useful for low-to-medium biomass and for change detection.
ALOS-2 PALSAR-2 (L-band): L-band SAR at 25 m (mosaic) or 3–10 m (spotlight). L-band penetrates deeper into canopy than C-band, raising the saturation threshold to roughly 150–200 Mg ha⁻¹ in some forest types, though saturation in dense tropical forest remains a known and unresolved problem. JAXA publishes annual global mosaics.

Why wall-to-wall maps need a sparse corrector

SAR-derived and optical-derived biomass maps are appealing precisely because they cover every hectare. The problem is that the relationship between backscatter or spectral reflectance and actual biomass is learned from field plots or allometric equations, both of which carry their own uncertainties. When that learned relationship is applied across a continent, small systematic errors compound into large ones. A map that is consistently 20 percent too high across a project area does not just misstate a stock figure; it corrupts every credit issuance that rests on it.

Spaceborne lidar does not solve this by mapping biomass directly. What it provides is a large, geographically distributed set of canopy height measurements that are physically grounded rather than statistically inferred. Height is then converted to biomass using allometric relationships, but the conversion is applied at the lidar footprint level rather than extrapolated across a biome. Those footprint-level estimates become the calibration dataset against which the wall-to-wall product is adjusted.

What GEDI actually delivers, and what it does not

GEDI's full-waveform returns decompose the vertical structure of a canopy in a way that photon-counting systems cannot fully replicate. The RH98 metric (the height below which 98 percent of waveform energy is returned) is a close proxy for top-of-canopy height and is the variable most commonly used in biomass allometries. Published studies using GEDI L2A and L2B products report canopy height RMSE values of roughly 2–5 m against field measurements in structurally complex tropical forests, rising in stands with high slope or dense understorey.

The honest constraint is geometry. GEDI's eight tracks are separated by about 600 m across-track. Within a single orbit pass, footprints are 60 m apart along-track. Over its operational period the ISS precession fills in coverage, but the result is still a sample, not a scan. In a 1,000 hectare project area you might have a few dozen usable GEDI shots on a good day, fewer after quality filtering for cloud contamination, urban proximity or steep terrain. The 51.6° latitude ceiling is absolute: Scandinavia, Canada north of roughly the 52nd parallel, and Siberia receive no GEDI data whatsoever. ICESat-2 fills that gap geographically, though its photon-counting approach yields different height metrics that require separate cross-calibration.

The calibration workflow in practice

The standard approach has three stages. First, quality-filter the lidar shots. For GEDI this means applying the l2_quality_flag, removing shots with beam sensitivity below 0.9, and masking slopes above roughly 30° where the waveform geometry becomes unreliable. For ICESat-2 ATL08, photon classification confidence thresholds and along-track terrain correction matter most. After filtering, shot counts in a typical project area drop substantially; this is expected and should be reported transparently in any MRV submission.

Second, convert filtered height metrics to biomass estimates using a published, peer-reviewed allometric equation appropriate to the forest type. Pantropical equations such as those from Chave et al. (2014) are commonly used, requiring wood density inputs from global databases. The resulting per-footprint biomass estimates carry uncertainty, which should be propagated rather than ignored.

Third, compare those footprint-level estimates against the co-located values in the wall-to-wall SAR or optical product. If the SAR product shows a consistent directional bias, a bias-correction factor or a recalibrated regression can be derived and applied. The corrected map is then the product submitted for carbon accounting. Crucially, the correction is only as good as the lidar sample density: where shots are sparse, uncertainty bounds widen and should be stated explicitly.

ICESat-2 as a boreal and high-latitude supplement

Above 51.6°, GEDI is absent and boreal forests hold a disproportionate share of global forest carbon. ICESat-2's ATLAS instrument, operating at 532 nm with six beams at 11 m diameter, reaches within a few degrees of the poles. Its ATL08 product segments canopy height at 100 m intervals and has been validated against airborne lidar in North American and Fennoscandian boreal forests with reported RMSE values generally in the 2–4 m range on flat to moderate terrain.

The cross-track spacing of ICESat-2 beams is narrower than GEDI's, but the along-track density is higher. The practical implication for calibration work is that ICESat-2 tends to perform better in elongated project areas aligned with the orbital ground track, and worse in compact areas where few tracks intersect. Neither sensor is a substitute for the other; a complete global biomass calibration programme uses both.

Uncertainty that must appear in the MRV record

Carbon registries increasingly require quantified uncertainty alongside biomass estimates. The lidar calibration workflow introduces uncertainty at several points: the allometric equation itself (often ±20–30 percent at the plot level), wood density assignments, slope correction in the lidar processing, and the spatial extrapolation from sparse shots to the wall-to-wall grid. None of these cancel neatly. A credible MRV submission states each source of uncertainty separately and propagates them through to a final confidence interval on the carbon stock figure.

SAR saturation is a separate, structural problem. In dense tropical forest above roughly 150–200 Mg ha⁻¹, L-band backscatter from PALSAR-2 ceases to respond meaningfully to additional biomass. Lidar calibration can identify where the map is in saturation territory, but it cannot recover information that the SAR never captured. In those stands, the honest position is a wide uncertainty bound rather than a point estimate. Satellize builds that uncertainty accounting into its biomass analytics outputs rather than reporting false precision.

For projects seeking registry-grade evidence packages, the calibration dataset itself, including the filtered lidar shots, the allometric conversion, the bias statistics and the corrected raster, should be archived and auditable. That is a data-management requirement as much as a technical one.

Typical figures

GEDI footprint diameter	25 m
GEDI along-track shot spacing	~60 m
GEDI across-track track spacing	~600 m (8 tracks)
GEDI latitude coverage	51.6° N to 51.6° S only
ICESat-2 ATLAS beam diameter	~11 m; 6 beams in 3 pairs
ICESat-2 ATL08 segment length	100 m along-track
ICESat-2 latitude coverage	Near-global to ~88°
Sentinel-1 SAR resolution (IW mode)	10 m; 6-day repeat at mid-latitudes
ALOS-2 PALSAR-2 L-band mosaic resolution	25 m; annual global mosaic from JAXA
L-band biomass saturation threshold	~150–200 Mg ha⁻¹ (forest-type dependent)

Analytics Satellize can run

Bias-corrected above-ground biomass raster	GEDI RH98 / ICESat-2 ATL08 height-to-biomass conversion via published allometry; ordinary least-squares or quantile regression against SAR wall-to-wall product	GeoTIFF raster at SAR native resolution with per-pixel uncertainty band
Lidar shot quality report	GEDI l2_quality_flag filtering, beam sensitivity thresholding, slope masking; ICESat-2 photon confidence filtering	PDF report listing shot counts before and after filtering, spatial distribution map, and flagged exclusion reasons
SAR saturation zone delineation	Identification of pixels where PALSAR-2 backscatter exceeds empirical saturation threshold relative to lidar-derived estimates	GIS polygon layer of saturation zones with associated uncertainty uplift values
Allometric uncertainty propagation table	Monte Carlo propagation of allometric equation error, wood density variance and lidar height RMSE through to per-stratum biomass confidence intervals	Spreadsheet and narrative annex suitable for registry submission
Cross-sensor height consistency check	Co-location of GEDI and ICESat-2 shots within shared coverage zones; Bland-Altman analysis of height metric agreement	Scatter plots, agreement statistics and recommendation on which sensor to weight per latitude zone
Calibration dataset archive package	Filtered lidar shot tables, allometric inputs, bias statistics and corrected raster assembled to audit trail standard	Versioned, timestamped data package in formats accepted by Verra, Gold Standard or equivalent registry

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.