Understory vegetation structure mapping with SAR penetration
L-band and P-band SAR penetrate closed forest canopies and return backscatter from shrub and herb layers that optical sensors never reach. Polarimetric decomposition separates what is canopy, trunk, and ground.
Sensors
- ALOS-2 PALSAR-2: L-band (1.27 GHz) quad-pol stripmap at 6 m resolution, 46-day exact repeat; full polarimetric data enables Cloude-Pottier and Freeman-Durden decomposition to isolate volume, double-bounce and surface scattering components beneath the canopy.
- ESA BIOMASS: P-band (435 MHz) fully polarimetric SAR; longer wavelength penetrates even dense tropical canopies more deeply than L-band, reaching trunk bases and ground layer. Operational acquisition began 2025; global coverage planned at roughly 25-day repeat.
- Sentinel-1 (C-band): C-band (5.4 GHz) IW mode at 10 m resolution, 6-day repeat over most landmasses. Penetration is limited to upper canopy in closed tropical forest, but useful for boreal and temperate stands where canopy is thinner, and for multi-temporal coherence analysis.
- TanDEM-X: X-band interferometric SAR pair; phase centre sits near the top of the canopy rather than the ground, so differencing TanDEM-X phase height against a terrain model gives canopy height context that anchors the sub-canopy decomposition products.
Why the canopy is not the forest
Optical sensors, including very high-resolution multispectral imagery, see what the sun sees: the uppermost leaf surface. In a closed-canopy tropical or boreal forest, that means a single layer of crowns, and everything below is invisible. The understory is not a footnote. Shrub density, herb cover, fallen woody debris and trunk basal area all govern carbon flux, fire behaviour, biodiversity and regeneration dynamics. None of it registers in a 30 cm RGB image.
SAR changes the geometry of the problem. Microwave energy at L-band (roughly 24 cm wavelength) passes through the leaf canopy with moderate attenuation and scatters from branches, trunks and the ground beneath. P-band (roughly 70 cm wavelength) goes further still, reaching trunk bases in dense tropical forest where L-band begins to saturate. The backscatter that returns encodes the vertical structure of the whole forest column, not just its roof.
What polarimetric decomposition actually separates
A single-pol SAR image gives you intensity. A fully polarimetric (quad-pol) image gives you the full 3x3 coherency matrix, which can be decomposed into physically meaningful scattering mechanisms. The two most widely applied frameworks are Cloude-Pottier eigenvector decomposition, which produces entropy (H), anisotropy (A) and mean scattering angle (alpha), and Freeman-Durden three-component decomposition, which separates volume scattering (dominated by randomly oriented branches and leaves), double-bounce (trunk-ground dihedral returns) and surface scattering (relatively smooth ground or water).
For understory mapping, the double-bounce component is the most diagnostic. A high double-bounce return beneath a volume-scattering canopy signals the presence of vertical stems at ground level, whether shrubs, regenerating saplings or standing dead wood. The ratio of volume to double-bounce power shifts measurably as understory density changes, even when the canopy above remains closed. ALOS-2 PALSAR-2 quad-pol stripmap data, archived from 2014 onward, is the primary operational input for this analysis. BIOMASS P-band data, now entering its acquisition phase, is expected to push detection sensitivity deeper into high-biomass tropical stands where L-band volume scattering saturates above roughly 100 t/ha.
Honest limits: what SAR penetration cannot resolve
The physics is real but the inversion is not clean. Volume scattering from the canopy and volume scattering from a dense shrub layer can produce similar polarimetric signatures, which is why decomposition alone rarely gives a unique answer. Ground-truth plots, ideally with measured basal area and shrub cover fraction at the time of acquisition, are needed to calibrate the model and validate class boundaries.
L-band saturates in high-biomass tropical forest. Above roughly 100 to 150 t/ha above-ground biomass, the signal is dominated by the canopy and trunk layer, and sensitivity to understory variation drops sharply. P-band extends this ceiling considerably, but BIOMASS data is new and calibration datasets are still being assembled by the science community. Steep terrain introduces layover and shadow artefacts that corrupt decomposition results in mountainous areas; slope correction and local incidence angle normalisation are mandatory preprocessing steps, not optional refinements. Finally, neither L-band nor P-band tells you species composition of the understory. Structure, yes. Floristics, no.
Building a time series: change in understory structure
A single-date decomposition product is a snapshot. The more useful product for land managers and conservation bodies is a multi-temporal stack that tracks how understory structure evolves after disturbance, selective logging, fire or restoration intervention. ALOS-2 PALSAR-2's 46-day repeat and the JAXA global mosaic programme, which has produced annual L-band mosaics since 2015, provide the archive depth to do this at national scale.
Change in the double-bounce component between annual mosaics has been used in published studies to detect the recovery of sub-canopy stem density after logging events, even when the canopy had already re-closed in optical imagery. This is the structural information that carbon verification schemes increasingly require: not just whether the canopy is present, but whether the forest column has recovered its full vertical complexity. Sentinel-1, despite its shallower penetration, contributes to this time series through coherence change detection, which flags disturbance events at 6-day intervals and flags where to focus the more computationally intensive quad-pol analysis.
From decomposition maps to operational products
The raw outputs of polarimetric decomposition are floating-point rasters of scattering power by component. Turning those into something a forest manager or carbon auditor can act on requires classification, validation and integration with ancillary data. A typical workflow chains terrain-corrected quad-pol backscatter through a Freeman-Durden or Cloude-Pottier decomposition, then feeds the component layers into a supervised classifier trained on field plots. The output is a map of understory structural classes, typically three to five classes ranging from open ground through sparse shrub to dense multi-layered understory, delivered as a GeoTIFF or vector polygon layer with per-class confidence.
Satellize runs this workflow on ALOS-2 PALSAR-2 archive data and, as BIOMASS acquisitions accumulate, will integrate P-band inputs for high-priority tropical sites. The company's Tonga crop-estimation programme demonstrates the same underlying principle of extracting agronomic structure from SAR backscatter in a data-sparse environment; the forest decomposition pipeline applies the same philosophy to a more complex scattering problem. For national forest inventory agencies or REDD+ project developers, the deliverable is a change-annotated structural map updated annually, with a validation report citing field-plot agreement statistics.
Choosing the right wavelength for your forest type
The decision between C-band, L-band and P-band is not a matter of preference. It follows directly from canopy closure and biomass density. In boreal forests with relatively open canopies and low leaf area index, Sentinel-1 C-band reaches the ground layer and decomposition is feasible. In temperate broadleaf forests with moderate closure, L-band from ALOS-2 PALSAR-2 is the standard choice. In dense tropical forest above 100 t/ha, P-band from BIOMASS is the only spaceborne option that consistently penetrates to the trunk base.
Airborne P-band systems such as ONERA's SETHI and NASA's AirMOSS have validated the physics in tropical and boreal environments over the past two decades, which is why the BIOMASS mission was designed around P-band rather than L-band. Those airborne datasets also serve as a useful calibration bridge while the BIOMASS archive is still shallow. For any procurement decision, the first question is not which sensor is most capable in the abstract, but which wavelength the target forest type actually requires.
Typical figures
| Primary SAR frequency | L-band 1.27 GHz (ALOS-2 PALSAR-2); P-band 435 MHz (ESA BIOMASS) |
| Spatial resolution (primary input) | 6 m (ALOS-2 PALSAR-2 quad-pol stripmap); ~50 m (BIOMASS baseline product) |
| Revisit period | 46 days exact repeat (ALOS-2); ~25 days (BIOMASS, target); 6 days (Sentinel-1 supplementary) |
| Polarisation modes required | Quad-pol (HH, HV, VH, VV) for full decomposition; dual-pol acceptable for simplified two-component separation |
| Canopy penetration depth | L-band: effective to ~20 m canopy depth in closed tropical forest; P-band: effective to trunk base in stands up to ~200 t/ha |
| Biomass saturation limit (L-band) | Approximately 100–150 t/ha above-ground biomass; sensitivity to understory degrades above this threshold |
| Archive depth | ALOS-2 PALSAR-2 from 2014; JAXA annual L-band mosaics from 2015; BIOMASS from 2025 |
| Delivery formats | GeoTIFF (decomposition component rasters, classified structural maps), GeoPackage (vector polygons), CSV (per-plot validation statistics) |
| Cloud sensitivity | None: SAR is cloud-transparent at L-band and P-band |
Analytics Satellize can run
| Understory structural class map | Freeman-Durden or Cloude-Pottier polarimetric decomposition followed by supervised classification (Random Forest or SVM) trained on field plots | Annual GeoTIFF with 3–5 structural classes and per-class confidence layer |
| Double-bounce power change layer | Multi-temporal differencing of Freeman-Durden double-bounce component across annual ALOS-2 PALSAR-2 mosaics | Change raster flagging gain or loss of sub-canopy stem density, updated annually |
| Volume-to-double-bounce ratio index | Per-pixel ratio of Freeman-Durden volume and double-bounce components, normalised by local incidence angle | Continuous index raster suitable for input to biomass or biodiversity proxy models |
| Post-disturbance understory recovery trajectory | Time-series decomposition over ALOS-2 archive (2014–present) aligned to known disturbance dates from Sentinel-1 coherence alerts | Per-stand recovery curve report with comparison against reference undisturbed stands |
| Terrain-corrected backscatter mosaic | Radiometric terrain correction using SRTM or Copernicus DEM, local incidence angle normalisation, multi-look averaging | Analysis-ready quad-pol GeoTIFF stack for client-side modelling |
| REDD+ sub-canopy structural verification layer | Integration of decomposition structural classes with above-ground biomass estimates to flag forest areas where canopy has re-closed but understory structure remains degraded | Verification polygon layer with confidence scores, formatted for carbon registry submission support |
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.