Standing dead wood and tree mortality mapping for carbon and biodiversity accounting

Standing dead wood is invisible to most forest inventories yet mandatory in IPCC Tier 2 carbon accounting. Red-edge indices, SWIR reflectance and spaceborne lidar waveforms can locate and quantify it, though confusion with drought-stressed live trees remains a genuine problem.

Sensors

Sentinel-2 MSI: 10 m visible and NIR bands, 20 m red-edge (bands 5, 6, 7) and SWIR (bands 11, 12); 5-day revisit at mid-latitudes with twin satellites. Red-edge chlorophyll indices and SWIR canopy-water indices are the primary spectral discriminators for dead wood, but 20 m pixels mix live and dead crowns in dense stands.
WorldView-3: 0.31 m panchromatic, 1.24 m multispectral, 3.7 m SWIR (8 bands, 1195–2365 nm); tasked commercial collection. The SWIR bands resolve individual crown reflectance anomalies associated with bark desiccation and absent foliar water, enabling single-tree mortality detection where Sentinel-2 cannot.
GEDI (NASA spaceborne lidar): Full-waveform lidar at 25 m footprint, ~60 m along-track spacing between footprints; ISS-based, covering 51.6° N/S. Dead canopy returns show reduced waveform amplitude, narrower vertical extent and altered relative height metrics (RH50, RH98) compared with live canopy of equivalent basal area. Useful for structural confirmation, not wall-to-wall mapping.
Landsat-8/9 OLI-TIRS: 30 m multispectral including SWIR bands 6 and 7; 100 m thermal (TIRS). 16-day single-satellite revisit. The thermal band can flag anomalously warm canopy surfaces associated with low transpiration in dead stands, though at 100 m the signal is heavily mixed. The archive back to 1972 (Landsat 1–9 series) supports long-term mortality chronology.

What a dead crown gives away spectrally

A living tree canopy absorbs strongly in red and blue wavelengths, reflects sharply in NIR, and absorbs again in SWIR because liquid water in foliage attenuates at 1.4 and 1.9 µm. When a tree dies, chlorophyll degrades within weeks to months, collapsing the red-edge inflection. Foliar water disappears faster still, lifting SWIR reflectance markedly. Bark, dry wood and lichen then dominate the signal.

The most diagnostic indices are red-edge chlorophyll index (CIre, using Sentinel-2 bands 7 and 5) and the normalised difference water index (NDWI or NDII, using NIR and SWIR). Dead standing wood typically shows CIre values below 1.5 and NDII values below roughly minus 0.1, though exact thresholds vary by species and decay stage. These are not universal cut-offs; they need site-level calibration against field plots or very-high-resolution imagery.

The drought-stress confusion problem

Severely drought-stressed live trees shed leaves, reduce stomatal conductance and drop foliar water content. Their spectral signature can closely mimic early-stage dead wood, particularly in seasonally dry forests during the dry season. This is the dominant source of commission error in automated mortality mapping.

Two strategies reduce it. First, multi-date compositing: a genuinely dead tree shows persistent low CIre and high SWIR reflectance across the following growing season, whereas a stressed-but-recovering tree rebounds. Second, thermal evidence from Landsat TIRS: dead canopy lacks transpirational cooling, so daytime land-surface temperature is measurably elevated relative to live neighbours, though at 100 m resolution this only works for patches of several dead trees, not isolated snags.

Neither approach eliminates the problem entirely. In forests where bark-beetle kill and drought stress co-occur, such as western North American conifers, omission and commission errors of 15–30% in pixel-level classifications have been reported in peer-reviewed studies. Honest inventory design should treat remotely sensed mortality maps as stratification layers for targeted field sampling, not as direct substitutes for ground measurement.

What lidar waveforms add that spectral indices cannot

GEDI full-waveform lidar records the vertical distribution of canopy material within each 25 m footprint. Live closed-canopy forest produces a broad waveform with strong upper and lower canopy returns. Dead standing wood, particularly in post-disturbance snag fields, produces a narrower waveform with reduced amplitude at upper canopy height and a stronger ground return, because the crown volume is reduced and foliage absent.

GEDI relative height metrics, particularly the ratio of RH50 to RH98, shift measurably in snag-dominated plots. This structural signal is independent of spectral chlorophyll content, which makes it a useful second line of evidence when spectral ambiguity is high. The limitation is spatial: GEDI samples along narrow tracks, leaving large gaps between swaths. It cannot produce wall-to-wall dead-wood maps on its own; it works best fused with Sentinel-2 spectral layers to extrapolate sampled structural observations across continuous spectral strata.

Feeding IPCC Tier 2 necromass carbon pools

IPCC Good Practice Guidance distinguishes five forest carbon pools: above-ground biomass, below-ground biomass, dead wood, litter and soil organic carbon. Dead wood is further split into standing dead wood (snags) and downed dead wood. Tier 2 accounting requires country-specific emission factors and pool estimates rather than global defaults, which means national forest inventories need spatially explicit dead-wood data.

The conversion from detected dead-wood area or crown count to carbon mass requires species-specific wood density values and volume-to-carbon conversion factors. These are published in the IPCC 2006 Guidelines and subsequent 2019 refinement. A practical workflow extracts dead-wood patch area from Sentinel-2 classification, estimates average stand density from GEDI or airborne lidar where available, applies species-appropriate density and decay-class corrections from field plots, and propagates uncertainty through the carbon estimate. The result is a necromass carbon pool figure with a defensible confidence interval, not a single point estimate.

Biodiversity accounting adds a parallel requirement. Standing dead wood is a keystone microhabitat: primary cavity nesters, saproxylic beetles and wood-decay fungi depend on snag density and size-class distribution. Emerging biodiversity accounting frameworks, including the Taskforce on Nature-related Financial Disclosures metrics, are beginning to require habitat-quality proxies. Snag density derived from satellite data is a credible proxy where field survey is impractical.

WorldView-3 SWIR for individual-crown resolution

Where a client needs single-tree mortality counts rather than patch-level area estimates, WorldView-3 SWIR is the only current spaceborne option. Its eight SWIR bands between 1195 and 2365 nm resolve individual crown reflectance at 3.7 m, sufficient to distinguish dead crowns from live neighbours in open-canopy forests and woodland savannahs. In closed-canopy tropical or boreal forest, crown overlap limits detection to emergent dead trees.

The practical constraint is cost and coverage. WorldView-3 is a tasked commercial system; large-area surveys are expensive, and cloud cover over tropical forests can frustrate collection windows for months. A tiered approach works well: Sentinel-2 at 20 m identifies candidate mortality patches across the full area of interest, and WorldView-3 is tasked selectively over those patches for crown-level validation. Satellize applies this tiered architecture in analytics work, drawing on open constellations for area-wide stratification and adding commercial tasking on client licence where higher resolution is warranted. The Tonga crop-estimation programme used a similar open-plus-commercial logic at the national scale.

Archive depth matters for mortality chronology. Sentinel-2 data runs from 2015; Landsat extends to 1972. Reconstructing when a mortality event began, and whether it is ongoing or stabilised, requires multi-year time series analysis. Breakpoint detection algorithms applied to NDII or NBR (normalised burn ratio, which also responds to canopy water loss from non-fire causes) can date mortality onset to within one or two growing seasons.

Honest limits and what field work still has to do

Satellite-based dead-wood mapping has a hard lower bound on detectable target size. A single snag of 30–40 cm diameter in a closed forest is invisible to Sentinel-2 and detectable by WorldView-3 only if it is an emergent tree above the surrounding canopy. GEDI footprints at 25 m average over many trees. This means fine-grained snag density, which ecologists measure in stems per hectare, cannot be derived from space alone.

Decay class is another gap. Standing dead wood passes through recognised decay stages from freshly dead to advanced structural collapse, each with different carbon density and biodiversity value. Spectral indices can distinguish fresh mortality (high SWIR reflectance, collapsed red-edge) from advanced decay (lower reflectance, partial crown loss), but the intermediate stages are spectrally ambiguous. Field verification of decay class remains necessary for high-confidence Tier 2 inventories.

What satellite data does well is stratify. A well-constructed mortality map from Sentinel-2 time series, validated with GEDI structural metrics, can reduce the field sampling effort needed for a national inventory by directing crews to high-probability mortality patches. That efficiency gain is real and quantifiable. The map is a stratification tool, not a replacement for the tape measure.

Typical figures

Spatial resolution (spectral)	10–20 m (Sentinel-2 MSI); 3.7 m SWIR (WorldView-3); 30 m (Landsat-8/9 OLI)
Spatial resolution (lidar)	25 m footprint, ~60 m along-track spacing (GEDI on ISS)
Revisit period	5 days (Sentinel-2 twin); 16 days (Landsat single satellite); GEDI non-repeating ISS orbit
Key spectral bands for mortality	Red-edge 705–783 nm (Sentinel-2 B5–B7); SWIR 1195–2365 nm (WorldView-3); SWIR 1566–2294 nm (Landsat OLI B6–B7)
Thermal band	100 m (Landsat TIRS, 10.6–12.5 µm); useful only for multi-tree patches, not single snags
Minimum detectable mortality patch	Approximately 0.1 ha at 20 m resolution (Sentinel-2); single emergent crown at 3.7 m (WorldView-3 SWIR)
Archive depth	Sentinel-2: 2015–present; Landsat: 1972–present; GEDI: 2019–present
Cloud limitation	Optical sensors fully obscured under cloud; multi-date compositing over 30–90 day windows partially compensates in temperate zones; tropical zones may have persistent gaps
Delivery formats	GeoTIFF classified raster, vector polygon (GeoJSON/Shapefile), carbon pool summary table (CSV), uncertainty bounds per stratum

Analytics Satellize can run

Dead-wood extent and patch map	Multi-date Sentinel-2 red-edge chlorophyll index (CIre) and NDII time-series breakpoint detection; persistent anomaly thresholding across two or more growing seasons	Annual GeoTIFF raster and vector polygon layer of mortality patches with onset date estimate and confidence class
Single-crown mortality count	WorldView-3 SWIR band ratio analysis at 3.7 m, applied to candidate patches identified by Sentinel-2 stratification; crown segmentation via object-based image analysis	Georeferenced point layer of detected dead crowns with species-group attribution where spectral library permits
Necromass carbon pool estimate	Dead-wood area from Sentinel-2 classification combined with GEDI RH metric-derived stand density; IPCC 2006/2019 wood density and decay-class conversion factors applied by stratum	Tabular carbon pool report by stratum with 90% confidence intervals, formatted for IPCC Tier 2 national inventory submission
Mortality chronology reconstruction	Landsat OLI NBR and NDII annual time series from archive (1972–present); breakpoint detection to date mortality onset by stand patch	Per-patch onset-year raster and summary CSV for disturbance history analysis
Drought-stress versus confirmed-mortality discrimination	Dual-season spectral composite (wet and dry season) combined with Landsat TIRS land-surface temperature anomaly; persistent versus transient signal classification	Revised mortality map with stress-confusion flags; recommended field-validation priority list by patch
Snag density biodiversity proxy	Mortality patch area and GEDI waveform amplitude reduction combined to estimate relative snag density class; cross-referenced against published saproxylic habitat thresholds from peer-reviewed forest ecology literature	Habitat-quality raster layer (snag density class 1–4) suitable for TNFD or national biodiversity strategy reporting

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.