Forest degradation detection from lidar waveform shape analysis
Spectral indices cannot see inside a canopy. Full-waveform lidar from GEDI and ICESat-2 exposes the vertical structure that degradation destroys first, making it the most direct spaceborne route to IPCC Tier 2 carbon accounting.
Sensors
- GEDI (Global Ecosystem Dynamics Investigation): NASA full-waveform lidar on the ISS. Circular footprints of approximately 25 m diameter, sampled at 60 m along-track intervals across eight parallel tracks. Delivers RH metrics (RH25 through RH98), plant area index profiles and waveform energy distribution. Coverage limited to latitudes between roughly 51.6°N and 51.6°S. No cloud penetration; acquisitions are clear-sky only. Archive runs from April 2019.
- ICESat-2 ATL08: NASA photon-counting lidar. ATL08 product classifies individual photons as ground, low vegetation or canopy, yielding canopy height percentiles and cover fraction at 100 m segments. Repeat cycle of approximately 91 days at the equator, tightening toward higher latitudes. Finer along-track sampling than GEDI but no waveform energy distribution; the two sensors are complementary rather than interchangeable.
- Sentinel-2 MSI: 10 m optical at 5-day revisit (two-satellite constellation). Supplies NDVI, EVI and red-edge indices as a spectral baseline against which lidar-detected structural change can be tested. The critical use here is demonstrating where spectral and structural signals diverge, which is the diagnostic fingerprint of degradation rather than deforestation.
- ALOS-2 PALSAR-2: L-band SAR at 25 cm wavelength. Penetrates canopy to the trunk-ground interface, providing backscatter proxies for above-ground biomass and detecting canopy gaps that optical sensors miss under cloud. Useful for corroborating GEDI-detected structural loss in persistently cloudy regions where lidar acquisition density is low.
What a waveform carries that a spectral index cannot
A lidar waveform is a time-series of returned energy from a single laser pulse. In an intact tropical forest the return is multi-modal: a first broad peak from the upper canopy, a secondary shoulder from mid-canopy foliage, and a sharp ground return. Selective logging, repeated understory burning or chronic drought stress each erode the mid-canopy peak while leaving the upper canopy largely intact. NDVI, which integrates reflected energy across the whole canopy column, sees almost nothing change. The waveform sees the collapse immediately.
GEDI quantifies this through relative height metrics. RH98 approximates the height of the tallest canopy elements. RH50 is the height below which half the total waveform energy is returned, a direct proxy for canopy bulk density. A degraded stand typically shows RH98 declining modestly while RH50 drops sharply, compressing the ratio. Plant area index profiles derived from the waveform add a further layer: they resolve foliage density at discrete height bins, making it possible to distinguish a canopy that has lost its understorey from one that has lost its emergent layer.
The metrics that matter and what they actually measure
RH75 minus RH25, sometimes called the waveform width or canopy relief ratio, is particularly sensitive to intermediate disturbance. Published work using GEDI L2A data has shown this metric separating logged-over forest from primary forest in Amazonian and Southeast Asian sites where spectral separation is statistically insignificant. RH98 alone is a poor degradation indicator because tall remnant trees can sustain a high value even when 60 percent of basal area has been removed.
Plant area index (PAI) profiles from GEDI L2B extend the analysis vertically. The L2B algorithm partitions waveform energy into discrete height bins and inverts a gap-fraction model to estimate foliage area density. A degraded stand loses PAI in the 5 to 20 m bin first. That bin corresponds to the sub-canopy layer where pioneer regrowth has not yet replaced removed timber. The signal is detectable at the individual footprint level but becomes statistically reliable when aggregated over 1 km grid cells, which is the standard unit for national forest monitoring.
ICESat-2 ATL08 contributes differently. Its photon-counting architecture resolves canopy height percentiles to finer along-track density than GEDI, and its 91-day repeat allows multi-temporal comparison at fixed ground tracks. Where a GEDI footprint is a single observation in time, repeated ICESat-2 passes over the same transect can detect progressive canopy lowering across seasons, which is the signature of chronic stress rather than acute disturbance.
Honest limits: what the physics and orbit geometry prevent
Both instruments are passive in one critical respect: they cannot fire through cloud. GEDI's acquisition is gated on clear-sky conditions, and in persistently cloudy tropical regions, such as much of the Congo Basin during the wet season, footprint density can fall well below the spatial sampling needed for reliable 1 km gridding. Published assessments of GEDI coverage over the tropics report that cloud screening removes between 30 and 70 percent of potential acquisitions depending on location and season.
The 25 m GEDI footprint is also not a map. It is a point sample. Interpolating between footprints to produce a continuous degradation surface requires either a statistical model trained on co-located field data or fusion with a wall-to-wall sensor such as Sentinel-2 or PALSAR-2. The interpolation introduces uncertainty that must be propagated honestly into any carbon accounting output. Claiming sub-footprint spatial precision from GEDI alone is not defensible.
A further ambiguity: waveform broadening can arise from terrain slope as well as canopy structure. A pulse fired onto a 20-degree slope returns energy across a longer time window purely because of geometry, mimicking the signature of a multi-layered canopy. GEDI's L2 products apply terrain corrections using the SRTM digital elevation model, but residual slope effects remain a known source of error in steep terrain. Any analysis in mountainous forest should treat slope-corrected metrics with appropriate scepticism and validate against field plots.
Connecting waveform metrics to IPCC Tier 2 carbon accounting
IPCC Tier 2 carbon accounting requires country-specific emission factors rather than global defaults. For the forest degradation category, that means estimating carbon stock change in forests that remain forests, the domain where waveform lidar has no spectral competitor. The IPCC 2006 guidelines and their 2019 refinement accept remote-sensing-derived estimates of carbon stock change provided they are calibrated against field measurements and uncertainty is quantified.
The operational pathway runs as follows. GEDI RH metrics and PAI profiles are regressed against co-located field measurements of above-ground biomass. The resulting model is applied across the full GEDI footprint archive for the country or jurisdiction. Footprint-level biomass estimates are then gridded and differenced between epochs to produce carbon stock change maps. Uncertainty bounds come from the regression error, the sampling density of footprints and the spatial interpolation method. Published studies using this approach in tropical forests report root-mean-square errors in the range of 20 to 40 Mg ha⁻¹ for above-ground biomass, which is acceptable for national-level Tier 2 reporting but requires transparent disclosure in the uncertainty table.
Satellize applies this workflow on open GEDI and ICESat-2 archives, adding PALSAR-2 backscatter as a gap-filling covariate in cloud-affected periods, an approach consistent with the fusion methods documented in the GEDI science team's published validation literature.
Where spectral and structural signals diverge: the diagnostic value of disagreement
The most analytically useful output of a combined lidar-optical workflow is not the degradation map itself but the disagreement layer: grid cells where Sentinel-2 spectral indices indicate stable or recovering forest while GEDI waveform metrics show structural decline. That pattern is the spaceborne fingerprint of cryptic degradation, the category most likely to be omitted from national greenhouse gas inventories and most likely to be exploited by actors who understand that optical monitoring has a blind spot.
Quantifying the extent of this disagreement across a jurisdiction gives forest agencies a defensible argument for investing in field verification in specific locations rather than across an entire landscape. It also provides a basis for distinguishing degradation from natural gap dynamics: a single treefall gap produces a localised waveform anomaly without a coherent spatial pattern, while selective logging produces clusters of anomalies aligned with terrain and access routes, a distinction that PALSAR-2-derived road detection can corroborate.
Typical figures
| GEDI footprint diameter | ~25 m; footprints spaced 60 m along-track, ~600 m across-track between beams |
| GEDI latitudinal coverage | 51.6°S to 51.6°N (ISS orbital inclination limit) |
| GEDI archive depth | April 2019 to present (with operational gaps) |
| ICESat-2 ATL08 segment length | 100 m along-track; repeat cycle ~91 days at equator |
| Sentinel-2 spatial resolution | 10 m (visible/NIR), 20 m (red-edge/SWIR); 5-day revisit at equator |
| ALOS-2 PALSAR-2 resolution | 3 m (spotlight) to 100 m (ScanSAR); L-band 1.27 GHz |
| Minimum detectable structural change (GEDI) | RH50 change of ~2 m detectable at footprint level; reliable at 1 km grid aggregation |
| Cloud penetration | None for GEDI/ICESat-2 (optical lidar); full for PALSAR-2 (SAR) |
| Biomass estimation RMSE (published range) | 20 to 40 Mg ha⁻¹ for tropical forest, regression-calibrated against field plots |
| Delivery formats | GeoTIFF waveform-metric grids, GeoPackage point clouds, CSV uncertainty tables, PDF Tier 2 accounting summary |
Analytics Satellize can run
| Waveform metric grid (RH50, RH75, RH98, PAI by height bin) | GEDI L2A/L2B product extraction, slope correction, 1 km gridding with footprint-density weighting | GeoTIFF raster stack, one band per metric, with footprint-count and uncertainty layers |
| Structural degradation classification | Threshold and random-forest classification on waveform metric ratios (RH50/RH98, PAI 5-20 m bin) trained against field-plot data | GIS polygon layer with degradation severity class (low, moderate, severe) and confidence score per 1 km cell |
| Spectral-structural disagreement map | Cell-wise comparison of Sentinel-2 NDVI/EVI trend against GEDI RH50 trend; flagging cells where optical signal is stable but structural signal declines | GeoTIFF divergence layer with flagged cells prioritised for field verification |
| Multi-temporal canopy height change (ICESat-2) | ATL08 canopy height percentile differencing across repeat passes on fixed ground tracks; Mann-Kendall trend test for progressive decline | CSV transect report with height-change magnitude, trend significance and seasonal decomposition |
| Above-ground biomass carbon stock change (Tier 2) | GEDI-field regression model applied across footprint archive; epoch differencing; uncertainty propagation per IPCC 2019 refinement guidelines | National or jurisdictional PDF accounting summary with uncertainty table, plus supporting GeoTIFF stock-change map |
| SAR-lidar fusion gap-fill for cloud-affected regions | PALSAR-2 HV backscatter used as biomass covariate to interpolate GEDI footprint estimates in low-density acquisition zones | Gap-filled GeoTIFF biomass surface with provenance layer distinguishing lidar-derived from SAR-interpolated cells |
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.