Parametric flood insurance trigger verification by SAR inundation mapping

SAR backscatter change detection maps flood extent objectively and near-real-time, giving parametric insurers a physics-based trigger that removes the adjuster from the critical path. This page covers the method, its honest limits, and the sensor options available today.

Sensors

Sentinel-1 A/B (C-band SAR, ESA): 5 x 20 m ground range resolution in Interferometric Wide Swath mode, 250 km swath. Repeat pass every 6 days at the equator with one satellite; 12-day cycle with the remaining operational unit after Sentinel-1B failure in 2021. Free and open data, typically available within 1–3 hours of acquisition via Copernicus Dataspace. The workhorse for large-area flood mapping.
ICEYE X-band SAR constellation: Sub-metre resolution in spotlight mode; stripmap mode delivers roughly 3 m resolution with a 30 km swath. Commercial tasking allows same-day revisit over a specific area of interest, which is critical for capturing peak inundation before recession. X-band is more sensitive to surface roughness and vegetation moisture than C-band.
Capella Space X-band SAR: Spotlight mode achieves 0.5 m resolution. Rapid tasking and delivery, with latency from collection to product often under 60 minutes. Useful for urban flood mapping where fine structural detail matters, though urban layover and shadow artefacts remain a challenge at any resolution.
ALOS-2 PALSAR-2 (L-band SAR, JAXA): L-band (1.2 GHz) penetrates vegetation canopies significantly better than C- or X-band, making it the preferred choice for detecting shallow inundation beneath forest or tall crops. 3–10 m resolution in high-resolution modes; 14-day repeat. Archive depth back to 2014 supports baseline construction.

Why open water betrays itself to radar

SAR sensors transmit microwave pulses and measure the energy scattered back to the antenna. Calm open water acts as a specular reflector: energy bounces away from the sensor rather than back to it, producing a sharp drop in backscatter relative to the surrounding dry land. The contrast is reliable enough that automated thresholding algorithms can delineate flood boundaries from a single post-event image, though change detection against a pre-event baseline reduces false positives from permanent water bodies, dark soils and road surfaces.

The physics does not require sunlight, does not care about cloud cover, and works at night. For flood insurance, these properties matter enormously. Optical sensors go blind precisely when tropical cyclones and monsoon depressions are producing the floods that generate claims. SAR does not.

From backscatter to a signed trigger

A parametric flood policy replaces subjective loss adjustment with a pre-agreed index: if inundation exceeds a defined area or depth within a defined polygon by a defined date, a payout is triggered automatically. SAR inundation mapping supplies the index value. The workflow is straightforward in principle. A pre-event backscatter composite is constructed from multiple passes to suppress noise. When a flood event begins, each new acquisition is co-registered to the composite, and pixels showing a backscatter decrease beyond a calibrated threshold are classified as inundated. The resulting binary or probabilistic flood mask is intersected with the policy polygon and the inundated area is computed.

Several published methods exist for the thresholding step. Simple histogram-based approaches work well over flat agricultural terrain. Bayesian change-detection frameworks, including the widely used Kittler-Illingworth method applied to SAR, reduce commission errors in heterogeneous landscapes. The Copernicus Emergency Management Service (CEMS) has used Sentinel-1-based flood mapping operationally since 2012, and its published activations provide a public benchmark for method validation.

Latency is the contractual variable that matters most. With Sentinel-1, acquisition-to-product latency of three to six hours is achievable using near-real-time processing chains. Commercial X-band operators such as ICEYE can deliver tasked imagery and derived flood masks within two to four hours of collection. For a parametric product, the insurer and insured agree in advance which sensor, which processing chain, and which latency standard constitutes the authoritative measurement.

Four limits you should price into the contract

Dense vegetation canopy is the most significant source of missed inundation. C-band and X-band signals are largely scattered by the canopy before reaching the water surface beneath. Floodplain forests, rice at late vegetative stage, and tall maize can all mask shallow flooding almost completely. L-band PALSAR-2 penetrates better but does not eliminate the problem. If the insured asset sits beneath a closed canopy, the policy trigger should either specify L-band data or acknowledge a detection floor, typically inundation depths below roughly 0.3–0.5 m under dense canopy may not register reliably.

Urban areas introduce layover, foreshortening and shadow artefacts that can mimic or obscure inundated pixels. Double-bounce between floodwater and vertical structures can actually increase backscatter, the opposite of the open-water signature, leading to underestimation of urban flood extent. Dedicated urban flood mapping algorithms exist but add complexity and uncertainty.

Wind roughens open water, reducing the specular effect and raising backscatter. At wind speeds above roughly 5–7 m/s, the clean low-backscatter signature of floodwater degrades, particularly for large open expanses. Shallow turbulent flow over fields can look similar to wet bare soil.

Revisit frequency is the timing risk. A flood may peak and partially recede between satellite passes. With a single Sentinel-1 satellite on a 12-day repeat, a fast-receding flash flood in a small catchment could be missed entirely. Commercial tasking solves this but adds cost. Contracts should specify a minimum observation frequency and define what happens if no cloud-free (for optical) or no valid SAR acquisition falls within the peak window.

Constructing a defensible baseline

The pre-event composite is not a trivial input. Seasonal variation in soil moisture, crop phenology and standing water from irrigation all shift the background backscatter level. A baseline built from a single dry-season image will produce false positives when compared against a wet-season acquisition even without a flood. Best practice uses a multi-temporal median composite from the same calendar period in prior years, typically three to five years of archive, to suppress seasonal noise.

Sentinel-1 archive depth extends to 2014 for most of the globe, giving a decade of reference data. ALOS-2 archive runs from 2014; its predecessor ALOS PALSAR from 2006 to 2011 extends the L-band record further. For regions with high inter-annual variability, a longer baseline reduces the probability of a spurious trigger, which matters as much to the insured as to the insurer.

What the product actually looks like

The deliverable for a parametric trigger is not a map for its own sake. It is a signed, timestamped measurement: inundated area in hectares within policy polygon X, derived from sensor Y, acquisition Z, processed by method W, with an uncertainty estimate. The uncertainty estimate is the part most vendors omit. A credible product states the commission and omission error rates expected for the land-cover type and sensor combination in use, drawn from published validation studies rather than invented.

Satellize structures its flood analytics outputs as GeoTIFF inundation masks with accompanying JSON metadata, including acquisition time, sensor, processing method version, and a per-polygon area summary suitable for direct ingestion by an insurer's claims system. The approach is consistent with how Satellize handles data pipelines in its other analytics work, including the Kingdom of Tonga crop-estimation programme, where objective, auditable measurement chains are a contractual requirement.

For reinsurance and catastrophe bond structures, the same SAR-derived flood extent can feed into an event footprint that is compared against an industry exposure database. The SAR measurement is sensor-agnostic in principle: the contract specifies the acceptable sensor list and the processing standard, and any conforming acquisition can serve as the trigger observation.

Typical figures

Spatial resolution (Sentinel-1 IW mode)	5 x 20 m ground range
Spatial resolution (ICEYE / Capella spotlight)	0.5–1 m
Swath width (Sentinel-1 IW)	250 km
Revisit (Sentinel-1, one satellite)	12 days at equator; 1–3 days at mid-latitudes with orbit geometry
Revisit (commercial X-band, tasked)	Same-day to next-day over a specific AOI
Acquisition-to-product latency	3–6 hours (Sentinel-1 NRT); 1–4 hours (commercial operators)
Frequency bands available	C-band 5.4 GHz (Sentinel-1); X-band 9.6 GHz (ICEYE, Capella); L-band 1.2 GHz (ALOS-2)
Minimum detectable inundated patch	Approximately 0.5–1 ha in open terrain at Sentinel-1 resolution; smaller with commercial SAR
Archive depth	Sentinel-1: 2014 to present; ALOS-2: 2014 to present; ALOS PALSAR: 2006–2011
Delivery formats	GeoTIFF flood mask, GeoJSON polygon, per-polygon area summary (JSON), WMS feed

Analytics Satellize can run

Binary flood-extent mask	SAR backscatter change detection with Kittler-Illingworth or Otsu thresholding against multi-temporal baseline composite	GeoTIFF raster, one layer per acquisition, timestamped
Per-polygon inundated area measurement	Zonal statistics applied to flood mask within insurer-supplied policy boundary GeoJSON	JSON report with area in hectares, acquisition metadata, and uncertainty band
Flood-onset and peak-extent timing estimate	Time-series analysis across all available acquisitions during event window, fitting inundation growth and recession curve	CSV time series plus summary report identifying probable peak acquisition
Canopy-masking risk assessment	Land-cover classification (ESA WorldCover or equivalent) intersected with policy polygon to flag high-vegetation fractions where C-band detection is unreliable	Risk-score layer and written caveat appended to trigger report
Multi-sensor consensus flood map	Ensemble combination of Sentinel-1 C-band and ALOS-2 L-band masks using majority vote or probabilistic fusion	Probabilistic inundation raster with per-pixel confidence score
Historical flood frequency baseline	Retrospective processing of Sentinel-1 archive (2014 to present) to produce annual maximum flood-extent composites for a defined polygon	GeoTIFF stack and summary table of annual inundated-area statistics for actuarial use

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.