Forest carbon flux attribution to disturbance type

Fire, logging, windthrow and drought mortality each release carbon at different rates and leave different spectral fingerprints. Multi-sensor fusion lets analysts attribute the cause, not just record the loss.

Sensors

Landsat 8/9 OLI: 30 m multispectral, 16-day revisit per satellite (8-day combined). Shortwave-infrared bands 6 and 7 (1.57–2.29 µm) are the primary discriminators: fire scars show high SWIR reflectance and low NBR; logging exposes bare soil with a distinct SWIR-NIR ratio; drought stress appears as a gradual NBR decline before structural collapse. The Collection 2 archive runs from 1982 (Landsat 4/5 TM) and from 2013 for OLI, giving trajectory depth no other free archive matches.
Sentinel-1 C-band SAR: 10 m resolution in IW mode, 6-day revisit at mid-latitudes, 12-day near the equator. VV and VH backscatter respond to canopy structure and moisture. Logging and windthrow produce abrupt coherence loss and a drop in VH backscatter of roughly 3–6 dB as canopy volume scattering disappears. Fire leaves a different signature: drier, lower-biomass residue can actually increase backscatter temporarily before it falls. Cloud-independent, which is decisive in humid tropical regions where optical revisit is effectively far longer than the nominal 8 days.
GEDI (NASA spaceborne lidar): 25 m footprint full-waveform lidar, approximately 25 m along-track spacing between shots. Operates from the ISS at roughly 51.6° inclination, covering latitudes to about ±52°. Provides canopy height and vertical structure metrics (RH50, RH98) that distinguish standing dead wood (drought mortality, bark beetle) from physical removal (logging) or flattening (windthrow). Repeat coverage of the same footprint is sparse and irregular, so GEDI is best used for before/after comparisons over multi-year windows rather than rapid change detection.
MODIS Terra/Aqua: 250–500 m resolution, near-daily global coverage. The 8-day surface reflectance composites (MOD09A1/MYD09A1) and the MODIS burned-area product (MCD64A1) provide coarse but temporally dense context for attributing large fire events and tracking drought-driven NDVI anomalies at landscape scale. Useful for constraining the timing of disturbance onset before finer-resolution sensors confirm the type.

Why the cause of death changes the carbon accounting

A hectare of dead forest is not a single entry in a carbon ledger. Fire combusts biomass immediately, releasing CO2, methane and nitrous oxide within hours to weeks. Logging removes merchantable timber and leaves slash that decomposes over years, with a portion of the carbon sequestered in wood products for decades. Windthrow kills trees but leaves them largely intact on the ground, where decomposition is slower than combustion and faster than product storage. Drought-induced mortality often leaves standing dead wood that decays over five to twenty years, with decomposition rates governed by local temperature and moisture.

These distinctions matter for national greenhouse gas inventories, REDD+ crediting and corporate supply-chain disclosures. Misattributing a logging event as fire, or a drought mortality patch as windthrow, produces systematic errors in flux estimates that compound across millions of hectares. The satellite record now carries enough spectral, structural and temporal information to do better than simple detection.

What a floating roof gives away: the spectral logic of attribution

Each disturbance type leaves a characteristic trajectory in the spectral indices derived from Landsat OLI. The Normalised Burn Ratio (NBR = (NIR - SWIR2) / (NIR + SWIR2)) drops sharply after fire and recovers within two to four years as pioneer vegetation establishes. Logging produces a similar NBR drop but with a distinct SWIR-to-red ratio that reflects exposed mineral soil rather than ash and char. The Normalised Difference Fraction Index (NDFI), developed for tropical forests, separates green vegetation, non-photosynthetic vegetation, shade and soil fractions and is particularly sensitive to the slash and exposed-soil signature of selective and clear-cut logging.

Drought mortality presents differently again. The NBR decline is gradual, often preceded by a multi-year reduction in the Enhanced Vegetation Index (EVI) as canopy water content falls. Landsat's SWIR band 6 (1.57 µm) is sensitive to liquid water in leaves; a slow reduction in that band's reflectance inversion, combined with stable or rising SWIR2 (2.29 µm), is a recognisable precursor. Windthrow is the hardest to separate spectrally from logging in a single image, because both expose soil and non-photosynthetic debris. The discriminating information comes from SAR and from the spatial pattern.

SAR backscatter and coherence as structural witnesses

Sentinel-1 C-band SAR adds two independent lines of evidence. First, the magnitude of VH backscatter change. Intact tropical forest typically returns VH backscatter in the range of roughly -10 to -7 dB in IW mode. Logging and windthrow remove the volume-scattering canopy layer, driving VH down by 3–6 dB depending on residual slash. Fire-affected forest shows a more variable response: the moisture content of the residue and the degree of combustion both influence whether backscatter rises or falls in the weeks immediately after the event.

Second, interferometric coherence. Coherence between two SAR acquisitions separated by 6 or 12 days is low in intact forest because the canopy moves between passes. After windthrow or logging, coherence rises as the scattering surface stabilises. After fire, coherence behaviour depends on whether standing structure remains. This coherence-magnitude combination, analysed in time series, provides a fingerprint that is genuinely complementary to optical indices rather than redundant with them. The combination is especially valuable in cloud-persistent regions where optical observations may be months apart despite the nominal 8-day Landsat revisit.

Height loss as the deciding variable

GEDI waveform lidar resolves the ambiguity that optical and SAR indices leave open. Logging removes canopy height; windthrow flattens it to ground level but leaves woody debris that produces a broad, low waveform return; drought mortality and bark-beetle kill leave canopy height largely intact while the RH50 metric (the height below which 50% of the waveform energy is returned) declines as foliage is lost from the upper canopy. Fire in closed-canopy forest typically collapses height and waveform energy together.

The practical limitation is GEDI's sparse and irregular revisit. A given 25 m footprint may be observed only a handful of times per year, and the ISS orbit means coverage is not guaranteed at a specific location within a short window. The workflow therefore uses GEDI as a validation and calibration layer rather than a primary change-detection trigger: optical and SAR time series flag the disturbance and narrow the candidate type, and GEDI footprints that fall within the disturbed patch confirm or revise the attribution. Where GEDI footprints are absent, canopy height models derived from GEDI-calibrated Landsat regression can substitute, with wider uncertainty bounds.

Translating attribution into flux estimates

Once disturbance type and extent are established, the carbon flux calculation follows published emission factor frameworks. The IPCC 2006 Guidelines and their 2019 refinements provide tier-1 and tier-2 emission factors by forest type and disturbance category. Fire emission factors incorporate combustion completeness, which varies between roughly 0.2 and 0.9 depending on forest type and fire intensity. Logging emission factors separate on-site combustion of slash, decomposition of residues and the carbon fate of harvested wood products. Windthrow and drought mortality emission factors are governed by decomposition rates, which are temperature and moisture dependent and therefore spatially variable.

The archive depth of Landsat Collection 2, extending to the early 1980s for TM data and continuously to the present for OLI, allows analysts to reconstruct prior disturbance history for a given stand. That history affects current emission estimates: a forest disturbed by logging twenty years ago carries less above-ground biomass than an undisturbed stand of the same age, so the emission from a subsequent fire is smaller. Ignoring prior disturbance history is a known source of overestimation in national inventories. Satellize's analytics pipeline for the Tonga crop-estimation programme uses a similar multi-date archive approach to establish baseline trajectories, and the same temporal logic applies here.

Honest limits: spatial resolution constrains the minimum attributable disturbance patch to roughly 0.1 ha for Landsat-only workflows and somewhat smaller when SAR is fused. Sub-pixel mixing is a real problem at forest edges and in selective-logging contexts where individual tree removal is below the detection floor. Drought attribution in particular carries higher uncertainty than fire or logging, because the spectral signal develops slowly and can overlap with phenological variation in seasonally dry forests.

What the output looks like in practice

The deliverable for a national forest monitoring programme is a disturbance-type map updated at a cadence matched to available cloud-free optical composites, typically quarterly in humid tropics and monthly in drier biomes, with SAR-based alerts at 12-day latency regardless of cloud. Each mapped polygon carries a disturbance-type classification (fire, logging, windthrow, drought mortality, or unattributed), a confidence score derived from the spectral-SAR-lidar evidence stack, and an associated carbon flux estimate with uncertainty bounds expressed in tonnes of CO2-equivalent per hectare.

Downstream users, whether national inventory agencies, carbon project verifiers or commodity-sourcing compliance teams, receive GIS layers compatible with standard forest monitoring platforms, accompanied by a methodology note that documents which sensor observations supported each attribution. Transparency about the evidence chain is not optional in carbon markets; it is the product.

Typical figures

Optical spatial resolution	30 m (Landsat 8/9 OLI); 250–500 m (MODIS, for temporal context)
SAR spatial resolution	10 m (Sentinel-1 IW mode, GRD product)
Lidar footprint	25 m diameter (GEDI full-waveform shots)
Optical revisit (nominal)	8 days combined Landsat 8+9; effective cloud-free revisit in humid tropics can exceed 60–90 days
SAR revisit	6 days at mid-latitudes, 12 days near equator (Sentinel-1 IW, single satellite)
Landsat archive depth	1982 (TM) to present; OLI from 2013; Collection 2 surface reflectance
Minimum attributable patch size	Approximately 0.1 ha for Landsat-SAR fusion; sub-0.1 ha with SAR alone at reduced confidence
Key spectral bands for attribution	SWIR1 (1.57 µm), SWIR2 (2.29 µm), NIR (0.86 µm) from OLI; VV and VH C-band (5.4 GHz) from Sentinel-1
Carbon flux output uncertainty	Typically ±20–40% at stand level, depending on disturbance type and biomass estimation method; fire lowest uncertainty, drought mortality highest
Delivery formats	GeoTIFF, GeoPackage, Shapefile, CSV flux tables; optional WMS/WMTS tile service

Analytics Satellize can run

Disturbance-type attribution map	Multi-index spectral trajectory analysis (NBR, NDFI, EVI) fused with Sentinel-1 VH backscatter change and coherence; random forest or gradient-boosted classifier trained on labelled disturbance patches	Quarterly GeoTIFF polygon layer with disturbance type, date of onset and confidence score per patch
Carbon flux estimate by disturbance type	Attributed disturbance extent multiplied by IPCC-tier emission factors by forest type and disturbance category; biomass density from GEDI-calibrated regression or published national maps	Tabular flux report (tCO2e per patch and aggregated by jurisdiction) with uncertainty bounds; delivered with methodology annex
SAR-based rapid disturbance alert	Sentinel-1 VH backscatter change detection against rolling 90-day baseline; coherence change layer as secondary flag; cloud-independent at 12-day latency	Fortnightly GeoJSON alert feed of candidate disturbance polygons, pre-classified by SAR signature type, for optical follow-up
GEDI-validated canopy height loss layer	GEDI RH98 and RH50 metrics compared across acquisition epochs within disturbed patches; gap-filled with Landsat-GEDI height regression where footprint coverage is absent	GeoPackage of height-change polygons with pre- and post-disturbance height estimates and attribution flag
Historical disturbance trajectory reconstruction	Landsat Collection 2 annual NBR composites back to 1984 processed through LandTrendr or CCDC change-detection algorithm to extract disturbance history per pixel	Per-pixel disturbance history raster (year, magnitude, duration) used to adjust current biomass and flux baselines
Drought mortality probability surface	Multi-year EVI and SWIR1 trend analysis against MODIS 8-day composites; anomaly detection relative to 20-year climatological baseline; cross-validated against GEDI waveform broadening in standing-dead patches	Annual raster of drought-mortality probability by forest stand, with flagged high-confidence patches for field verification

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.