Forest degradation and sub-canopy damage mapping

Selective logging, fuelwood extraction and edge erosion thin a canopy without clearing it. Binary deforestation alerts stay silent. L-band SAR and spectral unmixing reveal the gradient between intact and gone.

Sensors

ALOS-2 PALSAR-2: L-band (1.27 GHz) SAR with 3–10 m resolution in spotlight mode and 25 m in ScanSAR wide mode. L-band wavelength (~23 cm) penetrates the upper canopy and interacts with trunks and large branches, making it sensitive to structural changes that C-band misses. Repeat cycle 14 days; cloud-transparent.
Sentinel-2 MSI: 13 spectral bands, 10 m resolution in visible and near-infrared, 20 m in red-edge and SWIR. Red-edge bands (B5, B6, B7) are particularly sensitive to sub-lethal canopy stress and fractional photosynthetic cover changes. 5-day revisit at the equator with two satellites; severely limited by cloud in humid tropics.
Landsat 8/9 OLI: 30 m multispectral resolution across six reflective bands plus SWIR. The long archive (Landsat 5 onwards) enables phenology-normalised baselines essential for separating degradation signal from seasonal greening cycles. 16-day individual revisit, combined 8-day with both satellites operational.
GEDI (Global Ecosystem Dynamics Investigation): Spaceborne full-waveform lidar operating from the ISS. Provides canopy height profiles and foliage height diversity at ~25 m footprint, sampled in transects rather than full coverage. Directly measures vertical canopy structure loss that spectral indices can only infer. Coverage limited to roughly 51.6° N/S latitude.

Why degradation hides from standard change maps

Binary deforestation detection works by asking whether a pixel has switched from forest to non-forest. That question has a clean answer. Degradation asks something harder: how much forest remains, and how damaged is its structure? A selectively logged stand may retain 80 percent canopy cover. NDVI barely moves. The pixel stays green. The carbon loss and biodiversity impact, however, are substantial.

Selective logging in tropical forests typically removes only the largest commercial stems, leaving the canopy superficially intact but structurally impoverished. Fuelwood extraction removes understorey and sub-canopy layers that optical sensors cannot see at all. Edge degradation, driven by agricultural encroachment, degrades forest condition across a gradient that can extend hundreds of metres inward before any visible clearing occurs. None of these processes trigger a standard Hansen-style tree-cover-loss alert at the time they happen.

What L-band backscatter actually measures in a damaged stand

C-band SAR (Sentinel-1, 5.4 GHz, ~5.6 cm wavelength) interacts primarily with leaves and small branches. It is good at detecting clear-cuts but saturates at relatively low biomass densities, around 50–100 tonnes per hectare in many tropical forest types, and it cannot see through the canopy to structural changes below. L-band at 1.27 GHz has a wavelength of roughly 23 cm. That is long enough to pass through the leaf layer and scatter off woody trunks and large branches, the structures most affected by selective logging.

When large stems are removed, L-band HV (cross-polarised) backscatter decreases measurably. Published studies using ALOS PALSAR and its successor PALSAR-2 have demonstrated sensitivity to logging in Borneo, the Amazon and Central Africa, though the signal is subtle: changes on the order of 1–3 dB over logged areas, against natural spatial variability that can reach similar magnitudes. This is not a plug-and-play detection. It requires careful multi-temporal compositing, terrain correction and comparison against stable forest reference areas within the same scene.

Spectral unmixing and fractional cover: reading the canopy as a mixture

Rather than treating each Landsat or Sentinel-2 pixel as a single land-cover class, spectral unmixing decomposes the pixel into fractions of photosynthetic vegetation (PV), non-photosynthetic vegetation (NPV) and bare soil. The method, developed extensively for Australian savannas and later applied to tropical forests, is grounded in the spectral mixture analysis literature going back to the 1990s. In a degraded forest, the NPV fraction rises (dead wood, exposed litter) and the PV fraction falls, even when the canopy still looks green to a standard NDVI calculation.

The SWIR bands are critical here. Landsat OLI bands 6 and 7 (1.6 µm and 2.2 µm) are sensitive to dry vegetation and woody debris in ways that visible and near-infrared bands are not. Sentinel-2's 20 m SWIR bands offer similar capability with better spatial detail. Fractional cover time series, built from the Landsat archive back to the 1980s, can reveal degradation trajectories that would be invisible in any single-date image.

The phenology problem: separating damage from dry season

Seasonal forests are the most difficult environment for degradation mapping. In the dry season, deciduous and semi-deciduous species naturally shed leaves, the NPV fraction rises, NDVI drops, and L-band backscatter changes as the canopy dries. A naive change-detection algorithm reads this as degradation. It is not.

The standard mitigation is to compare observations at the same phenological phase across years, rather than adjacent dates within a year. This requires a long archive and careful date-matching. Landsat's 40-year record is genuinely valuable here. GEDI adds a complementary check: canopy height and foliage height diversity are structurally defined and far less sensitive to leaf phenology than spectral reflectance. Where GEDI transects cross an area of spectral ambiguity, the structural data can resolve whether the signal is seasonal or structural. The honest caveat is that GEDI does not provide wall-to-wall coverage; it samples in parallel transects spaced roughly 600 m apart at the equator, so it cannot substitute for area-wide mapping.

Combining the signals: a fusion approach and its honest limits

No single sensor solves this problem. The practical approach is to use Landsat fractional cover time series as the primary spatial layer (30 m, long archive, wall-to-wall), flag candidate degradation areas where PV fraction has declined outside the expected phenological envelope, then interrogate those candidates with PALSAR-2 backscatter change to confirm structural loss rather than spectral artefact. GEDI height profiles provide spot validation and help calibrate what a given backscatter change means in structural terms.

Minimum detectable degradation is genuinely uncertain. Published literature suggests that logging intensities below roughly 10–20 cubic metres per hectare are very difficult to detect reliably with current open-constellation sensors. At higher intensities, detection rates improve but false positives from phenological variation remain a real problem in seasonal forests. Cloud cover in humid tropical regions means that Sentinel-2 and Landsat observations may be months apart in practice, even with 5- and 8-day nominal revisits. SAR fills the temporal gap but introduces its own ambiguities.

Satellize applies this sensor fusion approach in its analytics work, building on the same open constellations used in the Tonga crop-estimation programme, where phenological normalisation against a multi-year baseline was equally central to producing reliable outputs.

What a degradation map can and cannot tell a buyer

A well-constructed degradation gradient map can tell you where canopy structure has been progressively impaired, at what approximate rate, and whether the pattern is consistent with selective logging, edge effects or fuelwood extraction based on spatial signature. It can support REDD+ jurisdictional accounting, concession compliance monitoring and conservation prioritisation in ways that binary deforestation alerts cannot.

It cannot reliably attribute cause without ancillary data. A drop in fractional cover and L-band backscatter is consistent with selective logging but also with storm damage, disease or drought stress. Ground truthing or very-high-resolution optical imagery is needed to confirm interpretation in ambiguous cases. Nor can it quantify timber volumes removed with any precision; the relationship between backscatter change and extracted volume is noisy and site-specific. Buyers who need legally defensible timber-volume estimates should treat satellite degradation mapping as a screening and prioritisation tool, not a substitute for field inventory.

Typical figures

Primary SAR spatial resolution	25 m (PALSAR-2 ScanSAR wide); 3–10 m in spotlight mode
Primary optical spatial resolution	30 m (Landsat OLI); 10–20 m (Sentinel-2 MSI)
SAR revisit (PALSAR-2)	14 days; cloud-transparent
Optical revisit (cloud-limited)	8 days combined Landsat 8/9; 5 days Sentinel-2 at equator; effective revisit in humid tropics may be 30–90 days
SAR frequency / wavelength	L-band 1.27 GHz / ~23 cm (PALSAR-2); C-band 5.4 GHz / ~5.6 cm (Sentinel-1, supplementary)
Optical spectral bands used	Red-edge (Sentinel-2 B5–B7), SWIR (Landsat B6–B7, Sentinel-2 B11–B12), NIR and visible
GEDI footprint and coverage	~25 m lidar footprint; transect sampling ~600 m apart at equator; 51.6° N/S latitude limit
Minimum detectable logging intensity	Approximately 10–20 m³/ha (published estimates; site-dependent)
Landsat archive depth	Landsat 5 onwards (~1984); Landsat 8/9 from 2013/2021
Delivery formats	GeoTIFF degradation gradient rasters, GeoPackage / Shapefile change polygons, time-series CSV by management unit

Analytics Satellize can run

Degradation gradient map	Spectral mixture analysis (PV/NPV/bare fractions) on Landsat/Sentinel-2 time series, phenology-normalised against multi-year baseline	Annual or biannual GeoTIFF raster classifying degradation severity (intact, lightly degraded, moderately degraded, severely degraded) per 30 m pixel
L-band structural change layer	Multi-temporal PALSAR-2 HV backscatter differencing with terrain correction and stable-forest normalisation	GeoTIFF of backscatter change magnitude overlaid on degradation gradient map; flagged candidate areas as GeoPackage polygons
Phenology-corrected NDVI and fractional cover time series	Same-phase annual compositing from Landsat archive; harmonic time-series decomposition to separate trend from seasonal signal	Per-pixel time-series CSV and summary charts by concession or administrative unit; trend direction and rate of change
GEDI-calibrated canopy structure assessment	GEDI Level 2A/2B canopy height and foliage height diversity intersected with SAR and optical degradation candidates	Tabular report of canopy height distribution within flagged degradation zones; used to validate or discount spectral candidates
Edge-effect degradation buffer analysis	Spatial gradient analysis of fractional cover decline from forest boundary inward; distance-decay modelling	GIS layer showing degradation depth from each forest edge, updated annually; summary statistics per concession boundary
Selective logging pattern classification	Spatial signature analysis (gap size, spacing, track geometry) on high-resolution optical and SAR data; comparison with known logging spatial patterns	Classified polygon layer distinguishing probable selective-logging signatures from storm damage and agricultural encroachment; confidence scores per polygon

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.