SAR L-band backscatter above-ground biomass estimation
L-band SAR backscatter correlates with above-ground biomass up to a saturation ceiling of roughly 100–150 t/ha, making it a practical but bounded tool for forest carbon accounting. Calibration against lidar or field inventory is not optional.
Sensors
- ALOS-2 PALSAR-2: L-band (1.27 GHz) SAR operated by JAXA. Stripmap mode delivers 3–10 m resolution; ScanSAR wide-area mode reaches 100 m. The annual global forest/non-forest mosaic, produced at 25 m, is the most widely used freely available L-band biomass input dataset.
- ESA BIOMASS (P-band, planned): P-band (435 MHz) SAR mission selected under ESA's Earth Explorer programme, designed to penetrate deeper into forest structure than L-band and push the backscatter saturation limit to 200–300 t/ha in some forest types. Launch is targeted for 2025.
- Sentinel-1 (C-band): C-band (5.4 GHz) limits canopy penetration to the uppermost leaf layer. Useful for detecting forest disturbance and monitoring open woodland, but backscatter saturates well below 50 t/ha in dense closed-canopy forest, making it a poor primary sensor for high-biomass estimation.
- GEDI (spaceborne lidar): NASA's Global Ecosystem Dynamics Investigation fires 1064 nm laser pulses from the ISS, sampling full waveform returns at 25 m footprints along eight parallel tracks. It does not produce wall-to-wall coverage, but its canopy height and vertical structure retrievals are the standard calibration anchor for SAR-based biomass models.
Why wavelength determines what the radar actually measures
Microwave backscatter from a forest is not a single signal. It is the sum of scattering from leaves, branches, woody stems and the soil beneath, weighted by how far the radar wavelength penetrates before being absorbed or scattered back. C-band (roughly 5 cm wavelength) interacts mainly with leaves and small twigs. Most of the energy never reaches the trunk. L-band (roughly 23 cm) passes through the leaf layer and interacts strongly with branches and woody stems, which is exactly where most above-ground biomass resides. P-band (roughly 70 cm) goes further still, reaching large stems and even the soil-trunk interface.
The practical consequence is that L-band backscatter in HV polarisation (horizontal transmit, vertical receive) shows a meaningful correlation with above-ground biomass across a useful range of forest densities. That correlation is not linear and it does not hold indefinitely. In dense tropical forest, the signal saturates: once the canopy closes and stems become large, additional biomass produces diminishing change in backscatter. The saturation level depends on forest type and structure, but published studies using ALOS PALSAR data consistently place it in the range of 100 to 150 tonnes per hectare for tropical and boreal closed-canopy forest. Below that ceiling, L-band is genuinely informative. Above it, you are largely guessing.
What the ALOS-2 PALSAR-2 global mosaic actually gives you
JAXA produces an annual 25 m global forest/non-forest mosaic from ALOS-2 PALSAR-2 acquisitions, normalised for acquisition geometry and made freely available. The HH and HV backscatter layers in that mosaic are the most practical L-band input for regional biomass mapping. They are consistent in processing, cover the tropics and boreal zone annually, and have an archive extending back to the original ALOS PALSAR mission from 2006, giving a time series of nearly two decades.
Turning those backscatter values into biomass numbers requires a retrieval model. The standard approach fits a water-cloud model or a similar semi-empirical function to paired backscatter and reference biomass observations, then applies it spatially. The reference observations come from field inventory plots, airborne lidar campaigns, or GEDI footprint retrievals. Without them, the backscatter-to-biomass conversion is poorly constrained. A model trained on one forest type will produce systematic errors if applied to another. This is not a software problem; it reflects genuine physical differences in forest architecture.
Soil moisture and topography: the two confounders that matter most
L-band backscatter responds to dielectric properties as well as structure. Wet soil and saturated vegetation increase backscatter independently of biomass. A forest measured two days after heavy rainfall will return higher HV values than the same forest measured in a dry period, even if nothing has grown. In seasonally flooded forests, the effect is amplified by double-bounce scattering between water surfaces and vertical trunks, which can produce backscatter signatures that mimic dense biomass. Acquisitions need to be screened for precipitation events, or multi-temporal compositing must be used to reduce moisture noise.
Topography introduces a separate problem. On slopes, the local incidence angle deviates from the nominal value, changing the path length through the canopy and altering the backscatter in ways that have nothing to do with biomass. Terrain correction using a digital elevation model is standard practice, but residual errors persist on steep slopes and in radar shadow zones. Mountainous forested terrain remains the hardest case for any SAR biomass retrieval, regardless of wavelength.
Where P-band changes the arithmetic
ESA's BIOMASS mission is specifically designed to address the saturation problem. At P-band frequencies, the longer wavelength interacts with large woody stems and the lower trunk, meaning backscatter continues to increase with biomass well beyond the L-band ceiling. ESA's published mission documentation cites a target of reducing uncertainty in above-ground biomass estimates in tropical forests to below 20% for areas larger than one hectare, with a design saturation limit substantially higher than L-band.
BIOMASS will also carry an interferometric mode intended to retrieve forest height through PolInSAR (polarimetric SAR interferometry), which provides structural information complementary to backscatter intensity. The mission is not yet operational, so published performance figures are pre-launch estimates. What is clear from the physics is that P-band offers a meaningful extension of the dynamic range for biomass retrieval in high-biomass forests, the very forests where carbon accounting uncertainty is currently largest.
Calibration is not a refinement. It is the method.
A SAR backscatter layer is not a biomass map. It becomes one only after calibration against independent observations that have their own uncertainty budgets. GEDI provides the most scalable calibration dataset currently available: its full-waveform lidar retrievals of canopy height and relative height metrics correlate strongly with field-measured biomass across multiple forest types, and its near-global sampling from the ISS means calibration points exist in most tropical and boreal regions. The limitation is spatial density. GEDI samples along discrete tracks; between tracks, the calibration relies on spatial interpolation and assumes forest type is consistent.
Field inventory plots remain the ground truth anchor. They are expensive, slow to collect, and sparse in the forests where biomass uncertainty is highest. The practical workflow for a credible L-band biomass product is: terrain-corrected PALSAR-2 HV backscatter, moisture-screened by acquisition date or multi-temporal compositing, calibrated against GEDI retrievals, with residual uncertainty quantified against available field plots. The output is not a single number per pixel but a distribution. Reporting biomass without an uncertainty estimate is not a minor omission; it makes the number unusable for carbon accounting purposes.
Satellize applies this workflow operationally, including for the Kingdom of Tonga crop-estimation programme where multi-source calibration is standard practice. The same principles apply at any scale from a national forest inventory to a project-level REDD+ monitoring area.
What you should and should not expect from an L-band biomass product
For forests below roughly 100 t/ha, a well-calibrated L-band product can achieve relative uncertainties in the range of 20 to 40% at the pixel level, improving significantly when aggregated to stand or landscape scale. That is useful for detecting deforestation, estimating carbon stocks in open woodland, and tracking regrowth after disturbance. For dense tropical forest above the saturation threshold, L-band alone cannot distinguish 150 t/ha from 300 t/ha. The radar simply does not see the difference.
Cloud cover is not a limitation for SAR, which is one of its principal advantages over optical methods in persistently cloudy tropical regions. Revisit frequency is a real constraint: the PALSAR-2 mosaic is annual, and while individual acquisitions can be tasked more frequently, L-band data is not available on the daily or weekly cadence that Sentinel-1 C-band provides. For change detection at high temporal resolution, C-band disturbance alerts remain the practical tool, with L-band providing the biomass baseline against which change is measured.
Typical figures
| Primary sensor frequency | L-band: 1.27 GHz (ALOS-2 PALSAR-2). P-band: 435 MHz (ESA BIOMASS, planned) |
| Spatial resolution (operational) | 25 m (PALSAR-2 annual mosaic); 3–10 m (PALSAR-2 stripmap); 25 m footprint (GEDI lidar calibration) |
| Revisit (PALSAR-2) | 14-day exact repeat; annual global mosaic product released by JAXA |
| Biomass saturation limit | ~100–150 t/ha (L-band, forest-type dependent); ~200–300 t/ha (P-band, pre-launch ESA estimates) |
| Polarisations used | HV (primary biomass signal); HH (forest/non-forest discrimination and moisture correction) |
| Cloud sensitivity | None. SAR operates through cloud and rain (some attenuation at C-band in heavy rain; negligible at L-band) |
| Archive depth (L-band) | ALOS PALSAR from 2006; ALOS-2 PALSAR-2 from 2014. Combined time series approaching 20 years. |
| Typical retrieval uncertainty (well-calibrated) | 20–40% relative at pixel level below saturation; improves to ~10–20% at stand or landscape scale with GEDI calibration |
| Delivery formats | GeoTIFF (backscatter and biomass layers), NetCDF (time series), GIS-ready shapefiles (change polygons), uncertainty rasters |
Analytics Satellize can run
| Above-ground biomass baseline map | Semi-empirical backscatter-to-biomass retrieval (water-cloud model or random forest regression) calibrated against GEDI canopy height retrievals and available field inventory data | GeoTIFF raster of estimated AGB in t/ha with co-registered uncertainty layer, delivered per project area |
| Annual biomass change detection | Multi-temporal PALSAR-2 HV differencing, thresholded against moisture-corrected baseline; confirmed against Sentinel-1 C-band disturbance alerts | Annual change polygon layer in GIS format, classified by loss/gain magnitude and flagged for likely cause (clearance, fire, regrowth) |
| Forest carbon stock estimate with uncertainty | AGB-to-carbon conversion (IPCC default factor 0.47 or species-specific where inventory data permits), aggregated to reporting unit with bootstrapped uncertainty propagation | Tabular carbon stock report by reporting unit, suitable for REDD+ or voluntary carbon registry submission, with methodology annex |
| Saturation-zone flagging | Backscatter variance analysis and GEDI-derived canopy height threshold applied to identify pixels likely above the L-band saturation ceiling | Confidence classification layer distinguishing retrievable-range from saturation-zone pixels, delivered alongside biomass map |
| Moisture-contamination screening | Acquisition-date precipitation filtering using ERA5 reanalysis or IMERG rainfall data; multi-temporal compositing to suppress wet-season anomalies | Quality-flagged backscatter composite with acquisition metadata, noting excluded scenes and compositing window |
| Terrain-corrected backscatter mosaic | Radiometric terrain correction using Copernicus DEM (30 m) or SRTM, applied per-scene before mosaicking; slope-shadow masking | Analysis-ready backscatter GeoTIFF mosaic at 25 m, with slope-angle and shadow mask layers included |
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.