Flood inundation extent and cropland damage assessment

SAR backscatter change detection maps flood extent over farmland within hours of a satellite pass, even through cloud. Intersecting that extent with crop-type maps and calendars converts pixels into tonnes of likely production loss.

Sensors

Sentinel-1 (ESA): C-band SAR at 5.405 GHz, 10 m ground range resolution in Interferometric Wide Swath mode, 250 km swath, 6-day repeat at the equator (3-day with both satellites). Free and open data policy makes it the operational backbone for flood mapping globally.
ALOS-2 PALSAR-2 (JAXA): L-band SAR at 1.27 GHz, 3–10 m resolution depending on mode. L-band's longer wavelength penetrates dense crop canopies far better than C-band, improving detection of inundation beneath standing rice or maize. Revisit is 14 days; data access requires a JAXA research agreement or commercial licence.
RADARSAT Constellation Mission (CSA): C-band SAR, three satellites providing 4-day exact repeat and sub-daily revisit at high latitudes. Compact polarimetry mode adds polarimetric information useful for distinguishing open water from flooded vegetation. Commercial tasking available.
Sentinel-2 MSI (ESA): 13 multispectral bands, 10–20 m resolution, 5-day revisit with both satellites. The SWIR band (Band 11, 1610 nm) and NIR are used to map surface water and assess post-flood vegetation stress via NDVI and NDWI, but cloud cover during active flood events severely limits utility. Most useful for pre-event crop-type mapping and post-flood damage confirmation.

Why open water looks almost black on radar

SAR sensors transmit microwave pulses and measure the energy scattered back to the antenna. Calm open water acts as a specular mirror: almost all energy reflects away from the sensor at the angle of incidence, and the return signal drops to near-noise level. Dry vegetated soil, by contrast, scatters diffusely and returns a strong signal. The contrast between flooded and non-flooded pixels is typically 3 to 8 dB in C-band VV polarisation, which is large enough to detect reliably with a simple threshold or a change-detection comparison against a pre-event reference image.

The practical consequence is that Sentinel-1 can produce a credible flood extent map within a few hours of a satellite pass, regardless of cloud cover or time of day. For agricultural damage assessment this matters enormously: optical sensors such as Sentinel-2 are often blinded for the entire duration of a flood event by the same convective cloud systems that caused the flooding. SAR is not.

The canopy problem: when crops hide the water underneath

The specular-reflection logic breaks down the moment a crop canopy closes over the floodwater. Dense rice, maize or sugarcane can produce a double-bounce return, where the microwave pulse reflects off the water surface and then off the vertical crop stem back to the sensor. This actually increases backscatter over flooded vegetated fields relative to the dry baseline, the opposite of what the open-water model predicts. Misclassification is common, and its magnitude depends on crop growth stage, row orientation relative to the look direction, and the degree of lodging caused by the flood itself.

C-band Sentinel-1 is most vulnerable to this effect because the 5.6 cm wavelength interacts strongly with centimetre-scale vegetation structure. L-band PALSAR-2 at 23.6 cm wavelength penetrates the canopy more deeply and reaches the water surface more reliably, making it the preferred sensor for flooded rice detection in vegetative growth stages. Published studies using PALSAR-2 have demonstrated detection of flooded rice fields at stages where Sentinel-1 would produce false negatives. Where L-band data are unavailable, analysts can partially compensate by using cross-polarisation (VH) alongside VV, and by incorporating phenological priors from the crop calendar to flag fields where canopy closure makes backscatter interpretation ambiguous.

From flood pixels to affected hectares: the intersection step

A flood extent raster is not itself a damage assessment. To convert it into something an agriculture ministry or food security agency can act on, the extent must be intersected with at least two additional layers: a crop-type map and a crop calendar.

The crop-type map assigns each field polygon or pixel to a specific commodity. Crop-type classification from multispectral time series is covered separately in this library; for flood damage work the key requirement is that the map be current-season, because planting patterns shift year to year. The crop calendar adds the critical dimension of growth stage at the time of inundation. A rice field flooded at transplanting may recover; the same field flooded at heading or during grain fill faces near-total yield loss. Flood duration compounds the damage: submergence tolerance thresholds for common varieties are well documented in agronomic literature, typically in the range of 3 to 14 days depending on variety and growth stage. Satellite-derived flood duration can be estimated by differencing the dates of the first inundation detection and the first post-flood clear observation, though this introduces uncertainty when revisit intervals are long or cloud persists.

The output of this intersection is an affected-area table by crop type and estimated growth stage, which can be fed into production-loss models using pre-event yield forecasts or historical statistics.

Latency, revisit and the operational window

Speed matters. Emergency food security assessments are most useful in the first 72 hours of a major flood event, when response agencies are still making decisions about relief logistics and governments are considering export restrictions or emergency imports. Sentinel-1's 6-day single-satellite revisit means that in the worst case a given location waits nearly six days for a usable pass. The two-satellite constellation brings this to roughly 3 days at mid-latitudes, and ascending plus descending orbit geometries can sometimes halve the effective wait further, though look-angle differences complicate direct comparison.

RADARSAT Constellation Mission's three-satellite design reduces revisit to 4 days globally and can be tasked for more frequent coverage over priority areas. For catastrophic events, the International Charter on Space and Major Disasters activates member agencies to provide priority tasking and data sharing, often delivering imagery within 24 hours of a request. Processing latency from raw SAR data to a georeferenced flood extent product is now typically 1 to 4 hours on cloud infrastructure, so the bottleneck is almost always satellite revisit rather than computation.

Honest limits: what the method cannot tell you

Several failure modes deserve explicit acknowledgement. Urban flooding is poorly handled: buildings produce strong double-bounce returns that mask inundation, and the method is generally not applied inside built-up areas. Very shallow flooding, less than roughly 5 to 10 cm, may not suppress C-band backscatter enough to cross detection thresholds, particularly over rough soil or short stubble. Wind roughening of open water surfaces raises backscatter and can cause missed detections; this is a known problem for large shallow lakes and flooded paddies in windy conditions.

Crop-type maps introduce their own uncertainty. If the map is derived from the previous season or uses a coarse classification, the affected-area estimates inherit that error directly. Production-loss estimates are further removed from observation: they depend on assumed yield levels, variety-specific submergence tolerance, and drainage speed, none of which are directly observable from orbit. The honest position is that satellite-derived flood damage assessments provide rapid, spatially consistent estimates of affected area, not audited loss figures. They are most valuable for triage and prioritisation, not for indemnity settlement (which is covered under a separate topic in this library).

Satellize applies this workflow operationally on Sentinel-1 and, where client licences permit, on commercial SAR tasking. The Tonga crop-estimation programme demonstrated how intersecting satellite-derived area estimates with crop calendars can produce actionable production figures for a national government operating without a dense ground-observation network.

Typical figures

Primary SAR spatial resolution	10 m (Sentinel-1 IW mode); 3–10 m (PALSAR-2, mode-dependent); 5–100 m (RADARSAT Constellation, mode-dependent)
Revisit interval	~3 days (Sentinel-1 two-satellite); 4 days (RADARSAT Constellation); 14 days (PALSAR-2)
Swath width	250 km (Sentinel-1 IW); 350 km (RADARSAT Constellation ScanSAR); 490 km (PALSAR-2 ScanSAR)
SAR frequency / wavelength	C-band 5.405 GHz / 5.6 cm (Sentinel-1, RCM); L-band 1.27 GHz / 23.6 cm (PALSAR-2)
Minimum detectable flooded area	Approximately 0.1 ha for open water in calm conditions at 10 m resolution; larger for flooded-under-canopy scenarios
Optical supplement resolution	10 m (Sentinel-2 Band 2, 3, 4, 8); 20 m (Sentinel-2 SWIR bands 11, 12)
Processing latency (SAR to flood extent)	1 to 4 hours on cloud infrastructure after data downlink and distribution
Sentinel-1 archive depth	From April 2014 (Sentinel-1A launch); global coverage available via Copernicus Data Space
Typical deliverable formats	GeoTIFF flood extent raster, vector polygon (GeoJSON / Shapefile), affected-area CSV by crop type, PDF situation report
Cloud sensitivity	SAR: none. Sentinel-2 optical: complete data loss under active storm cloud, limiting use to pre- and post-event windows

Analytics Satellize can run

Flood extent map	SAR backscatter change detection: thresholded difference between pre-event reference and flood-event image in VV and VH polarisation	GeoTIFF raster and vector polygon layer, delivered within 4 hours of SAR data availability
Flooded cropland area by crop type	Spatial intersection of flood extent with current-season crop-type classification layer	Tabular CSV and GIS layer showing affected hectares per commodity and administrative unit
Flood duration estimate	Multi-date SAR time series tracking inundation onset and recession across available passes	Per-field flood duration raster (days) with confidence interval based on revisit gap
Flooded-under-canopy detection	L-band PALSAR-2 backscatter analysis or C-band cross-polarisation ratio combined with crop phenology priors to identify double-bounce signatures over vegetated fields	Supplementary flood extent layer flagging canopy-masked inundation with associated uncertainty class
Post-flood vegetation stress assessment	Sentinel-2 NDVI and NDWI differencing between pre-event and first post-flood cloud-free acquisition	Stress severity map (GeoTIFF) and zonal statistics by field polygon
Estimated production loss	Affected area by growth stage combined with published submergence-tolerance thresholds and pre-event yield forecast; sensitivity range reported explicitly	PDF situation report with production-loss range by crop and province, suitable for food security briefings
Event alert	Automated SAR backscatter anomaly detection triggered when inundation exceeds a configurable area threshold within a defined agricultural zone	Email or API alert with preliminary affected-area figure, issued within hours of satellite 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.