Snow damage and windthrow detection in temperate forests
Catastrophic windthrow and snow-loading events can flatten thousands of hectares overnight. SAR coherence loss detects the damage within days regardless of cloud cover, while optical follow-up maps salvage priorities once conditions allow.
Sensors
- Sentinel-1 C-band SAR (ESA): 20 m resolution in Interferometric Wide Swath mode, six-day repeat at mid-latitudes (three days with both satellites). Cloud-penetrating, day-night capable. Coherence change between pre- and post-storm acquisitions detects canopy structural disruption within one repeat cycle. Backscatter increase in VV polarisation can indicate fallen stems on wet ground.
- Sentinel-2 MSI (ESA/Copernicus): 10 m resolution in visible and near-infrared bands, five-day repeat under cloud-free conditions. Used for damage-extent confirmation, species-level salvage prioritisation via the red-edge bands (bands 5, 6, 7 at 20 m), and multi-temporal monitoring of salvage-logging progress over the following months.
- TanDEM-X (DLR): Single-pass X-band SAR interferometry at 12 m resolution. Pre-event canopy height models derived from TanDEM-X provide the baseline against which post-storm height loss can be quantified, though new tasking is commercial and adds latency. Useful for severity stratification after the initial SAR alert.
- Planet SuperDove: 3 m resolution, eight spectral bands including red-edge, daily revisit over most land areas. Under clear skies, provides the highest spatial detail for mapping individual fallen-stem corridors and assessing the boundary between salvageable and unsalvageable timber. Commercial tasking; latency depends on licence.
What a fallen canopy looks like to a radar
When a storm flattens a forest stand, the electromagnetic environment at C-band changes in two distinct ways. First, coherence collapses. Interferometric coherence between a pre-storm and post-storm Sentinel-1 acquisition measures how similar the scattering geometry is between passes. Standing trees, even in wind, maintain a reasonably stable arrangement of branches and trunks. Fallen stems, displaced crowns and waterlogged debris do not. Coherence values that would typically sit above 0.6 in a mature conifer stand drop toward 0.2 or lower over heavily disturbed areas, a contrast that is straightforward to threshold and map.
Second, backscatter changes in a direction that depends on stem orientation and ground moisture. Stems lying parallel to the radar look-direction can produce a double-bounce return, increasing VV backscatter by several decibels. Stems perpendicular to the look-direction, or on saturated ground that absorbs rather than reflects, may produce a backscatter decrease. Neither response is universal, which is why coherence change is the more reliable primary indicator and backscatter change is treated as corroborating evidence rather than a standalone signal.
The six-day window that matters most
Salvage logging is economically viable for a limited period after a windthrow event. Bark beetles colonise fallen stems within weeks in warm conditions; once they do, timber value drops sharply and the biological calculus shifts toward leaving the material as deadwood habitat. Forest managers therefore need damage maps fast, not after the next cloud-free Sentinel-2 overpass, which in a Central European winter can be three weeks away.
Sentinel-1's six-day repeat at mid-latitudes, reduced to approximately three days when both Sentinel-1A and Sentinel-1B are operational, means a coherence-change product can be ready within one repeat cycle of the storm passing. The practical latency from storm landfall to a delivered damage map is typically five to eight days, dominated by the wait for the next SAR acquisition rather than processing time. That is still fast enough to inform initial salvage planning, road prioritisation and insurance assessment.
Snow-loading damage has a different temporal signature. Weight accumulation is gradual; the structural failure may be spread over days as temperatures fluctuate. SAR coherence responds to cumulative displacement, so a series of acquisitions across a loading event can track progressive damage rather than a single catastrophic break. This is worth building into the monitoring design when wet-snow events are forecast.
Honest limits of the method
Coherence change is sensitive but not specific. River flooding, agricultural tillage, and even heavy rainfall on bare soil can all suppress coherence over non-forest pixels, so masking with a pre-existing forest layer is essential before the analysis is meaningful. The Copernicus Global Land Service and the Hansen Global Forest Watch annual tree-cover products are the standard inputs for this masking step.
At 20 m resolution, Sentinel-1 cannot reliably distinguish individual fallen stems or small gaps below roughly 0.1 hectares. Scattered windthrow within a stand, where perhaps 20 per cent of stems are down but the canopy is not fully open, produces ambiguous coherence values that sit between the clearly damaged and clearly intact classes. Published studies on storms such as Kyrill (2007) and Lothar/Martin (1999) suggest detection rates above 80 per cent for contiguous damage patches larger than 0.5 hectares, with performance degrading for smaller or diffuse disturbances. Sub-hectare precision requires Planet or very-high-resolution commercial tasking.
TanDEM-X height-change analysis is powerful for severity quantification but the archive is not continuously updated; a pre-event acquisition from the global TanDEM-X DEM may be several years old, limiting its ability to capture canopy height changes that occurred before the storm.
Optical follow-up: from damage map to salvage plan
Once skies clear, Sentinel-2 adds detail that SAR cannot provide. The normalised difference vegetation index drops sharply over fallen or dying crowns, but the red-edge bands are more diagnostic: stressed conifers with needle browning show a characteristic red-edge shift that separates recently killed trees from bare soil or standing healthy canopy. This matters for salvage planning because it identifies stands where stems are down but crowns are still green, indicating recent damage and potentially recoverable timber.
Multi-temporal Sentinel-2 composites at monthly intervals after the event track salvage-logging progress spatially. Cleared areas show a spectral transition from damaged canopy to bare soil to, eventually, replanting signatures. This sequence is useful for verifying that salvage operations stay within the permitted damage footprint and do not expand into adjacent undisturbed stands, a compliance question that arises in nationally designated forests and Natura 2000 sites across Europe.
Putting the products to work
A practical monitoring workflow combines three layers: a rapid SAR coherence-change alert delivered within one repeat cycle of the storm, a confirmed damage-extent polygon once Sentinel-2 clears, and a monthly salvage-progress raster for the following six to twelve months. The alert layer is the operationally critical product; the optical layers support longer-term management and reporting.
Forest enterprises, reinsurers and national forest services are the primary users of this workflow in Europe. The damage-extent polygon feeds directly into timber-volume loss estimates when combined with a pre-event stand inventory or canopy height model. Satellize runs this kind of rapid disturbance analytics on open Sentinel data; the Tonga crop-estimation programme illustrates the same underlying logic applied to agricultural rather than forest canopies. The method transfers directly to any temperate or boreal forest where Sentinel-1 coverage is available, which is effectively the entire northern hemisphere land mass.
For events where speed is paramount, it is worth pre-positioning the analysis: setting up the coherence-change pipeline before storm season so that the first post-storm acquisition triggers an automated run rather than a manual one. A system that requires a human to remember to press go after a major storm is a system that will occasionally be slow when it matters most.
Typical figures
| Primary SAR resolution | 20 m (Sentinel-1 IW mode); 12 m (TanDEM-X) |
| SAR revisit at mid-latitudes | 6 days (Sentinel-1 single satellite); ~3 days (both satellites operational) |
| Optical confirmation resolution | 10 m (Sentinel-2 visible/NIR); 20 m (red-edge bands); 3 m (Planet SuperDove) |
| Optical revisit | 5 days (Sentinel-2 cloud-free); daily (Planet, weather permitting) |
| Radar frequency / band | C-band 5.4 GHz (Sentinel-1); X-band 9.6 GHz (TanDEM-X) |
| Minimum detectable damage patch | ~0.1 ha (SAR coherence threshold); ~0.5 ha for reliable detection above 80% rate in published studies |
| Typical alert latency post-storm | 5–8 days (dominated by SAR repeat cycle, not processing) |
| Sentinel-1 archive depth | From April 2014 (Sentinel-1A launch); enables pre-event baseline construction |
| Polarisations used | VV and VH (Sentinel-1 IW dual-pol); VV+VH coherence and backscatter ratio |
| Delivery format | GeoTIFF damage-extent polygons, GeoPackage or Shapefile; optional web map tile feed |
Analytics Satellize can run
| Rapid coherence-change damage alert | Interferometric coherence differencing between pre- and post-storm Sentinel-1 SLC pairs; threshold classification against forest mask | GeoTIFF raster and polygon layer of coherence-loss extent, delivered within 48 hours of post-storm SAR acquisition |
| Backscatter-change corroboration layer | Sigma-naught change detection in VV and VH polarisations; log-ratio of pre- and post-event calibrated backscatter | Supporting raster overlay indicating direction and magnitude of backscatter shift, flagging probable double-bounce zones |
| Confirmed damage-extent polygon | Sentinel-2 NDVI and red-edge index change detection against pre-storm baseline composite; fusion with SAR alert layer | Validated polygon shapefile with area statistics and severity class (partial vs. total canopy loss), produced once cloud-free optical acquisition is available |
| Timber-volume loss estimate | Damage-extent polygon intersected with pre-event canopy height model (TanDEM-X or national lidar) and stand-inventory attributes | Tabular report of estimated stem volume loss by stand compartment, for insurance or salvage-planning use |
| Monthly salvage-logging progress map | Multi-temporal Sentinel-2 compositing; spectral unmixing to track transition from damaged canopy to cleared ground to replanting | Monthly GeoTIFF raster series with cleared-area statistics; flags any expansion beyond permitted salvage boundary |
| Snow-loading progressive damage tracker | Time-series coherence analysis across multiple Sentinel-1 acquisitions during and after a wet-snow event; change-point detection on coherence trajectory | Animated raster series showing spatial progression of damage onset, with summary statistics per acquisition date |
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.