Wind-throw and storm-damage extent mapping using SAR coherence loss

Catastrophic wind events destroy interferometric coherence between SAR image pairs, leaving a measurable signal even under cloud. This page explains how C-band and L-band coherence loss is used to map storm damage extent, what the method can and cannot resolve.

Sensors

Sentinel-1 C-SAR (ESA): C-band (5.405 GHz), IW mode ground range detected resolution approximately 10 m range × 22 m azimuth, six-day repeat at mid-latitudes with two satellites, free open archive from 2014. The short wavelength makes it sensitive to upper-canopy disruption but limits penetration to the first few metres of canopy.
ALOS-2 PALSAR-2 (JAXA): L-band (1.2 GHz), stripmap mode at 3–10 m resolution, 14-day repeat. The longer wavelength penetrates the canopy and interacts with trunks and large branches, giving coherence loss that correlates more directly with stem displacement than surface litter disturbance.
RADARSAT-2 (MDA): C-band, quad-polarisation modes available at 8–25 m resolution, programmable revisit through commercial tasking. Quad-pol data support full polarimetric decomposition, which is the main analytical advantage over Sentinel-1's dual-pol IW mode.
Capella Space X-SAR: X-band (9.65 GHz), spotlight mode at sub-0.5 m resolution, tasked revisit within hours. X-band coherence is extremely sensitive to any surface change, useful for confirming damage boundaries at very high resolution, though the short wavelength means canopy penetration is negligible.

Why a fallen tree looks different to a radar

Synthetic aperture radar measures the phase and amplitude of backscattered microwave energy. A standing forest has a stable scattering geometry: the sensor sees roughly the same arrangement of vertical stems, branches and ground between repeat passes, so the phase difference between two acquisitions stays low and coherence stays high. When a storm flattens or tilts a stand, stems that were vertical are now horizontal, branch layers are collapsed, and the ground surface is exposed or covered in debris. The scattering geometry changes completely. Phase between the pre-event and post-event acquisition becomes random, and coherence drops sharply toward zero.

The depth of that coherence drop depends on wavelength. C-band (roughly 5 cm wavelength) interacts mainly with the upper canopy. Even partial crown loss or stem lean of 20 to 30 degrees is enough to destroy coherence. L-band (roughly 23 cm) penetrates further and responds to trunk displacement, making it more diagnostic of actual stem failure rather than crown stripping alone. Both are useful; they answer slightly different questions about the damage.

Turning a coherence map into a damage polygon

The standard workflow starts with a pre-event Single Look Complex (SLC) pair used to establish a baseline coherence value for the intact forest. A post-event SLC acquired as soon as possible after the storm is then co-registered to the pre-event image, and coherence is estimated over a multi-look window, typically 5 × 1 or 4 × 4 looks depending on the mode. The ratio of post-event to pre-event coherence, or simply the absolute post-event coherence, is thresholded to produce a binary damage mask.

Sentinel-1's six-day repeat is the practical workhorse here. In Europe, where Sentinel-1A and 1B historically provided that cadence, a major storm could be captured within days. Following the loss of Sentinel-1B in 2021, repeat intervals at some geometries extended to twelve days, though Sentinel-1C, launched in late 2023, is restoring the constellation. For events outside Europe, the repeat is longer and commercial tasking on RADARSAT-2 or Capella may be needed to close the temporal gap before vegetation regrowth or debris settling starts to re-stabilise coherence.

One honest complication: agricultural clearings, wet snow deposition and flooding also destroy coherence. A coherence-loss map is not automatically a wind-damage map. Ancillary data, storm track records, and optical imagery acquired when cloud clears are needed to confirm attribution.

What polarimetric decomposition adds

Coherence loss tells you where scattering geometry changed. Polarimetric decomposition tells you something about what the new geometry looks like. Decompositions such as Freeman-Durden or Cloude-Pottier separate the backscattered signal into surface scattering, volume scattering and double-bounce components. Wind-felled stems lying on the ground produce a strong double-bounce between the horizontal trunk and the exposed soil surface, a signature that is distinct from standing dead wood, which retains some vertical structure and continues to produce volume scattering.

This distinction matters for forestry agencies and insurers who need to separate total loss (stems down) from partial damage (crowns stripped, stems standing). RADARSAT-2 quad-pol data and ALOS-2 PALSAR-2 full-pol modes support this analysis. Sentinel-1 IW mode provides only dual polarisation (VV and VH), which limits decomposition to simpler two-component approaches. The separation is useful but not perfect: dense windthrow debris creates a complex mixed signal, and the double-bounce from wet soil and standing stems can be confused with that from downed trunks.

Resolution floors and what gets missed

The 10 to 20 m pixel size of most operational SAR modes sets a hard lower bound on what is detectable. A patch of wind-felled trees smaller than roughly 0.1 ha will be mixed with surrounding intact forest in a single resolution cell, and the coherence change will be diluted below any practical threshold. Individual tree losses, scattered gap openings, and narrow windthrow corridors along ridgelines are routinely missed. This is not a processing failure; it is a physical consequence of the wavelength and antenna size.

Very-high-resolution commercial SAR, such as Capella in spotlight mode, can push detection toward individual large stems, but at the cost of swath width and archive depth. For national-scale damage assessments after a major storm, Sentinel-1 coverage and archive depth outweigh the resolution penalty. For insurance-grade plot-level assessment, a combination of Sentinel-1 for rapid extent mapping and high-resolution optical or SAR follow-up for plot confirmation is the more practical approach.

Cloud cover, which is almost guaranteed in the aftermath of a major cyclone or extratropical storm, is not a problem for SAR. That is the single largest operational advantage of this method over optical damage mapping.

Operational timelines and archive value

Speed matters. Coherence is a difference measurement, and it degrades as time passes after the event. Debris settles, fallen stems begin to dry and change their dielectric properties, and secondary vegetation starts to colonise gaps. Studies of European windstorm damage using Sentinel-1 data suggest that coherence loss is clearest in the first one to three repeat cycles after an event, with signal quality declining thereafter. Getting the post-event acquisition quickly is more important than waiting for the ideal imaging geometry.

The Sentinel-1 archive, which runs from 2014, allows retrospective mapping of historical storm events. For countries building national forest disturbance records, this is significant: a decade of storm events can be reconstructed from open data, providing context for current damage and supporting long-term carbon accounting. JAXA's ALOS-2 archive adds L-band depth for tropical and subtropical regions where C-band coherence is complicated by dense, fast-regrowing vegetation.

Satellize has applied coherence-based change detection in the context of its broader satellite analytics work. For clients needing rapid post-storm damage extent, the workflow from SLC download to a georeferenced damage polygon layer can be completed within 24 to 48 hours of a suitable post-event acquisition becoming available.

Honest limits, stated plainly

SAR coherence loss is a strong indicator of structural change, not a direct measurement of timber volume lost or economic damage. Converting a damage polygon to a volume estimate requires stand inventory data or a biomass layer derived from separate methods. The method also cannot distinguish wind damage from ice storm damage, wet snow loading, or flooding in a single coherence image without supporting evidence.

In tropical forests, temporal decorrelation is fast even without disturbance, because the vegetation itself moves and grows between passes. Baseline coherence in intact tropical forest can be low enough that the contrast with damaged areas is compressed. L-band performs better than C-band in this environment, but neither is as clean as in temperate conifer stands. Analysts need to calibrate thresholds locally rather than applying a universal value.

Typical figures

Typical spatial resolution (Sentinel-1 IW)	10 m range × 22 m azimuth (GRD); ~5 × 20 m SLC before multi-looking
Typical spatial resolution (ALOS-2 PALSAR-2 stripmap)	3–10 m depending on mode
Revisit interval (Sentinel-1, single satellite)	12 days at equator; 6 days at mid-latitudes with two active satellites
Revisit interval (ALOS-2)	14 days
Frequency bands used	C-band 5.405 GHz (Sentinel-1, RADARSAT-2); L-band 1.2 GHz (ALOS-2); X-band 9.65 GHz (Capella)
Minimum detectable damage patch	Approximately 0.1 ha at 10–20 m resolution; smaller patches are below the reliable detection threshold
Post-event latency (open data)	24–72 hours from acquisition to Copernicus Data Space availability for Sentinel-1
Archive depth	Sentinel-1 from 2014; ALOS-2 from 2014; ALOS PALSAR (predecessor) from 2006
Cloud sensitivity	None. SAR penetrates cloud and rain at C- and L-band without signal degradation
Deliverable formats	GeoTIFF coherence ratio raster, binary damage mask shapefile, area statistics by administrative unit or forest compartment

Analytics Satellize can run

Post-event coherence ratio map	SLC co-registration and interferometric coherence estimation (multi-look window); pre/post coherence ratio normalised against intact-forest baseline	GeoTIFF raster at native resolution with coherence values 0–1, delivered within 48 hours of post-event SLC availability
Binary wind-damage extent polygon	Adaptive thresholding of coherence ratio map, morphological filtering to remove speckle artefacts, minimum mapping unit applied at 0.1 ha	Shapefile or GeoPackage of damage polygons with area (ha) and mean coherence-loss value per polygon
Damage severity classification	Three-class severity map (partial crown loss, severe crown loss, stem failure) derived from coherence magnitude and backscatter intensity change in VV and VH channels	Classified GeoTIFF with per-class area statistics in accompanying CSV
Polarimetric scattering decomposition layer	Freeman-Durden or Cloude-Pottier decomposition on quad-pol RADARSAT-2 or ALOS-2 data to separate downed-stem double-bounce from residual volume scattering of standing dead wood	Three-component RGB decomposition GeoTIFF and interpretation note distinguishing total stem loss from crown-only damage
Retrospective storm-damage time series	Coherence change detection applied systematically across the Sentinel-1 archive (2014 to present) for a defined forest area, events identified by storm-track dates from meteorological records	Annual damage-event catalogue as GeoPackage with event date, affected area (ha) and dominant severity class per event
Damage extent briefing report	Automated area statistics combined with analyst commentary on confidence, ambiguities and recommended optical follow-up acquisitions	PDF situation report suitable for forest agency or insurer, delivered within 72 hours of first post-event SAR pass

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.