Commercial SAR rapid flood extent mapping
Synthetic aperture radar maps flood extent in any weather, day or night, by detecting the near-zero backscatter of smooth open water. Commercial constellations now cut revisit from days to hours, but wind and forest canopy introduce real ambiguities that analysts must account for.
Sensors
- ICEYE SAR constellation: X-band (9.65 GHz), stripmap mode at approximately 3 m ground resolution, spotlight mode to sub-1 m. Constellation of more than 30 satellites as of 2024 enables revisit to any point on Earth within hours rather than days. ICEYE publishes flood-extent products directly and has supplied imagery under the Copernicus Emergency Management Service.
- Capella Space SAR: X-band, spotlight mode to approximately 0.5 m resolution, sliding spotlight and stripmap modes for wider area. Smaller constellation than ICEYE but same-day tasking is routine. Fine resolution aids urban flood boundary delineation where field boundaries are narrow.
- Sentinel-1 (ESA/Copernicus): C-band (5.405 GHz), Interferometric Wide Swath mode at 10 m resolution, 250 km swath. Free and open archive back to 2014. Nominal revisit 6 days with both satellites; reduced to 12 days after Sentinel-1B failure in 2021. Workhorse for baseline flood mapping but revisit is too slow for fast-onset events.
- COSMO-SkyMed (ASI): X-band, dual constellation (first and second generation) giving enhanced revisit. Stripmap at 3–15 m, spotlight to approximately 1 m. Operated by the Italian Space Agency; imagery available commercially and routinely activated under the International Charter: Space and Major Disasters.
Why smooth water goes dark on radar
SAR sensors illuminate the ground with microwave pulses and measure how much energy bounces back to the antenna. Rough surfaces, such as vegetation, buildings and churned soil, scatter energy in many directions; a significant fraction returns to the sensor, producing bright pixels. A calm water surface behaves differently. It acts as a specular reflector, bouncing almost all energy away from the sensor at the angle of reflection rather than back along the incident path. The result is near-zero backscatter: dark pixels in the SAR amplitude image.
This physical contrast is the foundation of SAR flood mapping. Analysts threshold the amplitude image, classify pixels below a backscatter level as open water, and compare the result against a pre-event baseline image of the same geometry to identify newly inundated areas. The method works at C-band (Sentinel-1 at 5.4 GHz) and X-band (ICEYE, Capella, COSMO-SkyMed at roughly 9–10 GHz) because the specular-reflection principle holds across both frequency ranges for water surfaces that are smooth relative to the radar wavelength.
Hours matter. Revisit arithmetic for disaster response
Sentinel-1 built the flood-mapping community its baseline capability, and its free archive is genuinely irreplaceable for pre-event reference imagery. The problem is time. With one satellite operational since 2021, the nominal 6-day repeat has stretched to 12 days over many regions. A flash flood that peaks and recedes in 48 hours may never be captured at its maximum extent by Sentinel-1 at all.
Commercial X-band constellations change that arithmetic. ICEYE's published constellation size, exceeding 30 satellites, supports revisit windows measured in hours for priority areas. Capella offers same-day tasking as a standard service. This is not a marginal improvement. For emergency managers deciding where to pre-position rescue assets, the difference between a 12-day repeat and a 4-hour repeat is the difference between a map and a historical document.
Latency after acquisition matters too. ICEYE publishes flood-extent vector products within hours of acquisition for activated events. Raw SAR amplitude data from commercial operators typically reaches analysts within 2–4 hours of downlink, depending on ground-station geometry.
Two ambiguities that will mislead an inattentive analyst
Wind roughening is the first. When wind speed exceeds roughly 3–5 metres per second over open water, surface capillary waves scatter enough microwave energy to raise backscatter above the calm-water threshold. A flooded field in a gusty storm can appear as bright as dry land. The effect is wavelength-dependent: X-band is more sensitive to small-scale roughness than C-band, so ICEYE and Capella imagery can be more susceptible to wind-induced false negatives than Sentinel-1 in the same conditions. Analysts must cross-reference wind observations or model output and treat wind-affected acquisitions with appropriate scepticism.
Dense forest canopy is the second. When floodwater sits beneath a closed tree canopy, the radar pulse interacts primarily with the canopy and the trunk-water double-bounce mechanism rather than the water surface directly. Double-bounce can actually increase backscatter over flooded forest relative to dry forest, which is the opposite of the open-water signature. Standard thresholding will miss inundation under closed canopy entirely, or misclassify it. This is a known and unresolved limitation: no single SAR acquisition reliably maps water beneath dense tropical forest. Multi-temporal change detection and polarimetric decomposition methods reduce the error but do not eliminate it.
Polarisation and what it adds
Most flood-mapping workflows use VV polarisation (vertical transmit, vertical receive) because smooth water suppresses VV backscatter most strongly. Sentinel-1 acquires VV and VH simultaneously in IW mode, and the ratio between them carries additional information. Flooded vegetation tends to show elevated VH relative to VV because of volume scattering and double-bounce, whereas open water suppresses both. Dual-polarisation data therefore supports finer class separation: open water, flooded vegetation and dry land can be distinguished with greater confidence than single-polarisation data alone allows.
ICEYE's standard stripmap product is single-polarisation (HH or VV depending on the acquisition). Sentinel-1's dual-pol IW mode remains the practical standard for polarimetric flood classification at scale, which is one reason the Sentinel archive remains central to the method even when commercial constellations provide the timeliness.
From amplitude image to actionable flood extent
Raw SAR amplitude data is not a flood map. Processing steps include multi-looking to reduce speckle noise, terrain correction using a digital elevation model (the Copernicus DEM at 30 m is the standard open reference), co-registration of the event image to a pre-event baseline, and backscatter change detection or direct thresholding. Automated pipelines can complete this chain in under an hour on cloud infrastructure; the bottleneck is usually data access latency rather than computation.
The output is typically a binary or three-class raster (dry, flooded, uncertain) that is then vectorised and intersected with exposure datasets: population grids, road networks, agricultural parcel boundaries, critical infrastructure locations. That intersection step is where the flood extent becomes operationally useful. A 10 km² flooded polygon is interesting; a count of how many people, roads and hospitals fall inside it is actionable.
Satellize runs this processing chain on both open Sentinel-1 data and commercially tasked X-band imagery for clients who need faster revisit than the open constellation provides. The crop-estimation programme in the Kingdom of Tonga has demonstrated the same underlying change-detection logic applied to agricultural monitoring, which shares analytical lineage with flood-extent delineation over agricultural land.
What the method cannot do
Flood depth is not directly measurable from SAR amplitude alone. Extent and depth are different quantities, and conflating them is a common mistake in briefings to non-specialist decision-makers. Depth estimation requires either in-situ gauge data, hydraulic model assimilation, or repeat-pass interferometry under very specific conditions. SAR gives you the shoreline of the flood, not the bathymetry beneath it.
Urban areas present a related problem. Buildings create complex multi-bounce geometries that can either mask flooded streets or produce bright double-bounce signatures that mimic dry urban structure. Very-high-resolution X-band spotlight imagery (sub-1 m from Capella) improves urban flood delineation considerably relative to 10 m Sentinel-1, but street-level inundation in dense city blocks remains difficult to map reliably from any spaceborne SAR system currently operating.
Typical figures
| Spatial resolution (commercial X-band stripmap) | Approximately 3 m (ICEYE, COSMO-SkyMed); sub-1 m in spotlight mode (Capella, ICEYE) |
| Spatial resolution (Sentinel-1 IW mode) | 10 m, 250 km swath |
| Revisit (commercial constellation) | Hours to same-day for priority tasking; ICEYE constellation exceeds 30 satellites as of 2024 |
| Revisit (Sentinel-1, single satellite) | 12 days at mid-latitudes (6 days when both satellites operational) |
| Radar frequency | C-band: 5.405 GHz (Sentinel-1); X-band: ~9.6–9.65 GHz (ICEYE, Capella, COSMO-SkyMed) |
| Typical data latency (commercial) | 2–4 hours from acquisition to analyst delivery; flood-extent products within hours for activated events |
| Cloud and daylight sensitivity | None. SAR is active microwave; cloud cover and solar illumination are irrelevant |
| Minimum detectable flooded area (open terrain) | Approximately 0.1 ha at 3 m resolution; larger at 10 m (Sentinel-1 practical floor ~1 ha) |
| Sentinel-1 open archive depth | 2014 to present; free access via Copernicus Data Space |
| Standard delivery formats | GeoTIFF amplitude, Cloud-Optimised GeoTIFF, GeoPackage or Shapefile flood-extent vectors |
Analytics Satellize can run
| Rapid flood extent raster | Backscatter thresholding and change detection against pre-event Sentinel-1 or commercial SAR baseline | Cloud-Optimised GeoTIFF, delivered within 4 hours of acquisition downlink |
| Flood extent vector with uncertainty class | Dual-polarisation (VV/VH) ratio classification with wind-flag masking; three classes: open water, flooded vegetation, uncertain | GeoPackage layer for direct import into GIS or emergency management platform |
| Exposure intersection report | Spatial join of flood-extent polygon against population grid (WorldPop or national census), road network and infrastructure point layers | PDF and machine-readable JSON summary: affected population count, road-km inundated, critical facilities inside flood boundary |
| Multi-date flood progression series | Time-series of co-registered amplitude images processed through consistent pipeline; change in flooded area quantified between acquisitions | Animated GIF and tabular CSV of flooded-area extent per acquisition date |
| Agricultural loss preliminary estimate | Intersection of flood extent with crop parcel boundaries and crop-type map; flooded parcel area aggregated by crop class | GIS layer and summary table of flooded area by crop type, suitable for insurance or government damage assessment |
| Wind-contamination quality flag | ERA5 or GFS wind-speed field co-located with acquisition time and footprint; pixels in wind-speed bands above 3 m/s flagged as low-confidence | Quality-flag band appended to flood-extent raster |
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.