Urban flood extent rapid mapping during and after events
SAR backscatter change detection maps open-water extent in flooded cities through the cloud cover that defeats optical sensors. Sentinel-1, COSMO-SkyMed and Capella Space provide the imagery; honest analysis requires acknowledging what radar shadow hides.
Sensors
- Sentinel-1 GRD (C-band, ESA): 10 m ground-range detected imagery, 250 km swath in IW mode, 6-day repeat at the equator (3-day with both satellites). Free and open. The workhorse for flood mapping globally; C-band penetrates cloud and light rain but is attenuated by very heavy precipitation.
- COSMO-SkyMed Second Generation (X-band, ASI): Spotlight modes reach 1 m resolution; StripMap delivers 3–5 m over a 40 km swath. Rapid-tasking turnaround of under 12 hours is available for emergency activations. X-band gives sharper urban geometry but is more sensitive to rain attenuation than C-band.
- Capella Space SAR (X-band, commercial): Spotlight imagery to approximately 0.5 m resolution, with same-day tasking commercially available. Small constellation limits revisit, but the resolution resolves individual flooded street segments that Sentinel-1 cannot separate.
- Maxar WorldView (optical, post-event): 30 cm panchromatic resolution for post-event damage confirmation once skies clear. Used to validate SAR-derived flood extents, identify debris lines and support building-damage triage. Cloud-dependent; not useful during the event itself.
Why radar is the only sensor that works when you need it most
Flood events and cloud cover are not coincidental companions. Convective storms, tropical cyclones and monsoon systems produce the precipitation that causes urban inundation and simultaneously generate the thick cloud decks that render optical satellites useless for days. A WorldView pass at 30 cm resolution is irrelevant if the city is under eight oktas of cumulonimbus.
Synthetic aperture radar illuminates the ground with its own microwave pulse and records the return signal. Cloud water droplets are orders of magnitude smaller than the wavelength of C-band (5.6 cm) or X-band (3.1 cm) radiation, so both pass through cloud with negligible loss under most conditions. Very heavy rainfall does attenuate X-band measurably, which is worth noting when interpreting COSMO-SkyMed or Capella data during the peak of a storm. C-band Sentinel-1 is more tolerant of precipitation, which partly explains its dominance in operational flood programmes.
What a flooded street looks like to a radar
Open water acts as a specular reflector: the radar pulse bounces away from the sensor rather than back to it, producing a characteristically dark return in the SAR image. Dry urban fabric, by contrast, produces bright returns from roads, rooftops and the double-bounce mechanism between vertical building walls and horizontal ground. Change detection compares a pre-event reference image, acquired under dry conditions, with the event or post-event acquisition. Pixels that were bright and have become dark are candidates for inundation.
The double-bounce effect that makes buildings bright in dry conditions also complicates flood detection. When water fills the space between a building wall and the ground, the geometry can actually increase the double-bounce return rather than suppress it, particularly in VV polarisation. Analysts use VH cross-polarisation, which is less sensitive to the double-bounce geometry, alongside VV to reduce this ambiguity. Permanent water bodies (rivers, canals, reservoirs) must be masked using a pre-event water layer derived from the same sensor or from a long-term surface-water dataset such as the JRC Global Surface Water product, otherwise the change signal is dominated by features that were always wet.
The parts of the city that radar cannot see
Honesty about radar geometry is not optional. SAR images from a side-looking geometry, and tall structures create two systematic artefacts: layover, where the top of a building appears closer to the sensor than its base, and shadow, where the ground behind a tall structure receives no illumination at all. In dense high-rise districts, shadow zones can cover a substantial fraction of the street-level area. Inundation behind a row of towers is simply invisible to the sensor.
Building density and orientation relative to the satellite track both affect how much ground is obscured. Ascending and descending passes from different look directions partially compensate for this: a street in shadow on an ascending pass may be illuminated on a descending pass. Sentinel-1 acquires both geometries over most land areas, and combining them reduces, though does not eliminate, the blind zones. For emergency response purposes, analysts should flag shadow-masked areas explicitly on delivered maps rather than presenting them as confirmed dry ground. This distinction matters when directing rescue resources.
Speed, resolution and the trade-off between them
Sentinel-1's 6-day repeat (3-day with both satellites active) is adequate for slow-onset river floods but can miss the peak of a fast urban flash flood. COSMO-SkyMed Second Generation and Capella Space accept priority tasking requests that can deliver imagery within hours of a request, at the cost of narrower swaths and, for most commercial operators, per-scene fees.
Resolution determines what the map can resolve. At 10 m, Sentinel-1 GRD can identify flooded city blocks and major road corridors but cannot reliably distinguish a flooded residential lane from a dry one if the lane is narrower than roughly 20–30 m (accounting for speckle filtering that further degrades effective resolution). Capella's sub-metre spotlight mode can resolve individual flooded driveways. The appropriate sensor depends on the operational question: a national emergency agency needs rapid wide-area coverage; a city drainage authority assessing a single district may justify a commercial tasking order.
The Copernicus Emergency Management Service activates Sentinel-1 flood mapping within hours of a formal request under the International Charter on Space and Major Disasters. Delivered products typically include grading maps that distinguish observed water, possibly flooded areas and areas not observed due to radar shadow, which is the correct disclosure standard.
From pixels to an operational product
A production flood-mapping workflow involves several steps beyond simple thresholding. Pre-processing converts raw GRD data to terrain-corrected sigma-naught backscatter. A reference composite, ideally a median of several dry-season acquisitions, is differenced from the event image. A threshold, often determined adaptively using the histogram of the difference image, separates flooded from non-flooded pixels. Shadow and layover masks derived from a digital elevation model are applied. Permanent water is subtracted. The result is a binary or graded flood extent layer, typically delivered as a GeoTIFF or vector polygon in a coordinate system the client's GIS can ingest directly.
Depth estimation from SAR alone is not reliably possible in urban areas. SAR measures surface extent, not volume. Hydraulic models constrained by the SAR-derived extent boundary can estimate depth, but that requires a calibrated model of the urban drainage network, which is a separate analytical product and is not always available for rapidly growing cities in lower-income countries. Satellize's analytics team can combine SAR-derived extents with available terrain data to produce depth-probability layers where model inputs exist; the Tonga crop-estimation programme gave the team direct experience integrating open satellite data with locally held ground datasets under time pressure.
After the water recedes: optical confirmation and damage triage
Once cloud clears, very-high-resolution optical imagery from Maxar WorldView or equivalent systems provides the confirmation layer that SAR cannot. At 30 cm, analysts can identify debris lines (which mark the high-water extent even after drainage), structural collapse, vehicles displaced by flow, and the sediment staining that marks inundated ground. This imagery feeds building-damage assessments and insurance loss estimation workflows.
The combination of SAR during the event and optical post-event is not redundant. SAR gives the emergency manager a map within hours; optical gives the recovery planner a damage inventory days later. Neither alone is sufficient. The archive depth of Sentinel-1, which extends to 2014, also allows analysts to compare current events against historical flood extents from previous years, supporting longer-term risk characterisation for urban planners and insurers.
Typical figures
| Spatial resolution (Sentinel-1 IW GRD) | 10 m (ground range), effective resolution after speckle filtering typically 20–30 m |
| Spatial resolution (COSMO-SkyMed SG Spotlight) | 1–3 m depending on mode |
| Spatial resolution (Capella Spotlight) | Approximately 0.5 m |
| Revisit (Sentinel-1, both satellites) | 6 days at equator; 3 days with ascending and descending combined; shorter at higher latitudes |
| Tasking latency (commercial SAR) | Same-day to next-day for priority orders; subject to constellation availability and conflict with other customers |
| Radar frequency | C-band (5.6 cm, Sentinel-1); X-band (3.1 cm, COSMO-SkyMed SG, Capella) |
| Minimum detectable flooded area | Approximately 0.5–1 ha for Sentinel-1 IW; smaller patches detectable with X-band at finer resolution |
| Archive depth (Sentinel-1) | 2014 to present; global coverage variable by year and region |
| Delivery formats | GeoTIFF (flood extent raster), GeoPackage or Shapefile (vector polygons), PDF summary map |
| Known blind zones | Radar shadow and layover behind tall structures; fraction of urban area affected depends on building height and satellite look angle |
Analytics Satellize can run
| Flood extent map with shadow mask | SAR backscatter change detection (sigma-naught differencing, adaptive thresholding) with DEM-derived layover/shadow mask | GeoTIFF and vector polygon layer showing flooded, possibly flooded and unobserved areas; updated per available pass |
| Permanent water subtraction layer | Pre-event SAR reference composite and JRC Global Surface Water mask applied to isolate new inundation from baseline water bodies | GIS layer of new inundation only, ready for emergency operations overlay |
| Multi-pass flood progression timeline | Time-series of SAR acquisitions during and after event, registered to common grid, showing advance and retreat of water extent | Animated GIF or multi-band GeoTIFF stack; tabular area statistics per time step |
| Depth-probability surface (where terrain model available) | SAR-derived flood boundary used as lateral constraint in terrain-based hydraulic approximation; uncertainty quantified from DEM error | Raster depth-probability layer with confidence intervals; flagged as model-dependent |
| Post-event optical damage triage | Very-high-resolution optical change detection (pre/post comparison) using Maxar WorldView or equivalent; debris line and structural collapse identification | Building-level damage classification layer (affected / possibly affected / no visible damage); PDF summary |
| Historical flood frequency layer | Sentinel-1 archive analysis (2014 to present) applying consistent change-detection algorithm across multiple historical events | Flood-frequency raster (number of events per pixel) for urban risk zoning and insurance exposure assessment |
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.