Flood exposure assessment for property asset valuation

Sentinel-1 SAR and Copernicus DEM hydraulic modelling can map flood inundation at the individual parcel level, quantifying 1-in-100-year and 1-in-200-year exposure for property asset registers. This page explains the physics, the method, and the honest limits.

Sensors

Sentinel-1 SAR (C-band, 5.4 GHz): Interferometric Wide Swath mode delivers 10 m ground range resolution across a 250 km swath. Cloud-penetrating C-band backscatter drops sharply over open water, enabling flood-extent mapping within 6 to 12 hours of an overpass. Revisit is 6 days at the equator and 1 to 3 days at mid-latitudes with both satellites.
Copernicus DEM GLO-30: Global 30 m (1 arc-second) digital elevation model derived from TanDEM-X radar interferometry. Absolute vertical accuracy is approximately 4 m RMSE globally, degrading to 6 m or worse in flat coastal and deltaic terrain where relief is smaller than the error budget. Used as the terrain input for hydraulic return-period modelling.
Landsat 8/9 OLI: 30 m multispectral imagery with a 16-day revisit. SWIR Band 6 (1.57–1.65 µm) and NIR Band 5 distinguish water from wet soil and vegetation during and after flood recession, useful for mapping slow-draining ponding that SAR can misclassify under wind-roughened conditions.
MODIS Terra/Aqua: 250 m to 500 m resolution with near-daily global coverage. Too coarse for individual parcel analysis, but the MODIS Near Real-Time Global Flood product (hosted at NASA FIRMS and LANCE) provides synoptic flood extent within 24 hours, useful for triggering higher-resolution SAR tasking.

What a SAR image sees when a field becomes a lake

Synthetic aperture radar works by measuring the strength of a microwave pulse reflected back to the satellite. Calm open water acts as a specular reflector: the pulse bounces away from the sensor rather than back to it, producing a characteristically dark return. Vegetated land, rooftops and roads scatter energy back strongly. The contrast is large enough that a simple change-detection comparison between a pre-flood reference image and a flood-event image produces a reliable open-water mask at 10 m resolution.

The Sentinel-1 archive stretches back to 2014, giving analysts a deep library of pre-event baselines for almost any location on Earth. During a major flood, the European Commission's Copernicus Emergency Management Service routinely publishes grading maps within 24 to 48 hours of a suitable overpass. Those maps follow a documented methodology: log-ratio thresholding of backscatter change, followed by morphological filtering to remove isolated dark pixels caused by calm ponds or shadow. The same approach is reproducible on any Sentinel-1 scene using open tools, which matters for auditability in a valuation context.

From a single flood event to a return-period probability

Knowing that a parcel flooded in October 2023 is useful. Knowing that it sits inside the 1-in-100-year inundation zone is what a lender, insurer or institutional investor actually needs. Translating observed extents into return-period probabilities requires a hydraulic model driven by a terrain surface.

The standard open approach combines Copernicus DEM GLO-30 with a 1D or 2D hydrodynamic model such as HEC-RAS or LISFLOOD-FP. Water surface elevations derived from satellite altimetry or gauge records are used to calibrate the model for observed events; the model is then run at design discharges corresponding to 1-in-50, 1-in-100 and 1-in-200 annual exceedance probabilities. The resulting inundation extents are intersected with a property asset register to assign each parcel a probability band and an estimated inundated area.

The intersection step sounds mechanical but contains real analytical choices. A parcel centroid inside a flood polygon is not the same as a parcel with its ground floor below the modelled water surface elevation. Depth matters for damage functions. Where building footprint data and floor-level surveys exist, depth-damage curves published by bodies such as the UK Environment Agency allow expected annual damage to be estimated as a monetary figure, which can then be discounted into an asset value adjustment.

The vertical accuracy problem in flat terrain

The Copernicus DEM GLO-30 has a published absolute vertical accuracy of approximately 4 m RMSE globally. In a river catchment with 20 m of relief across a floodplain, that error budget is manageable. In a low-lying coastal delta or reclaimed agricultural plain where the difference between a flooded and a dry parcel is 0.5 m of elevation, it is not.

This is the central honest limitation of any satellite-DEM-based flood model in flat terrain. A 4 m vertical error can shift a modelled flood boundary by hundreds of metres horizontally on a 0.1% slope. LiDAR-derived terrain models, where available, reduce vertical RMSE to 0.1 to 0.3 m and are strongly preferable for coastal and estuarine property portfolios. National LiDAR programmes exist for the UK, Netherlands, parts of the United States and a growing number of other jurisdictions. Where they do not exist, the satellite-derived result should be presented as a screening tool rather than a definitive exposure figure, and uncertainty bounds should accompany every parcel-level output.

SAR-observed flood extents from actual events partially sidestep this problem: they record where water actually went, not where a model predicts it would go. Building a multi-event archive of observed extents for a specific geography is often more reliable than relying solely on modelled return periods, particularly in areas where the DEM is known to be poor.

Integrating flood exposure into an asset register

A property portfolio might contain tens of thousands of parcels across multiple countries and climate zones. Running a bespoke hydraulic model for each geography is not practical. The operational approach is a tiered screening: first, intersect the entire register with existing published flood zone datasets and satellite-observed historical extents to identify high-exposure parcels; second, apply detailed modelling only to those flagging above a chosen risk threshold.

The output for each parcel typically includes the probability band (inside or outside the 1-in-100-year zone), the estimated inundated area and depth for each return period, the number of satellite-observed flood events in the archive, and a flag for whether the parcel sits in terrain where DEM uncertainty is high enough to warrant LiDAR verification. Delivered as a GIS layer joined to the asset register, this allows portfolio managers to sort by exposure, identify concentration risk in specific catchments, and prioritise physical surveys.

Satellize structures this kind of analysis as a reproducible pipeline: open Sentinel-1 and Copernicus DEM inputs, documented processing steps, and outputs joined to client-supplied asset registers. The same pipeline logic underpins the crop-estimation work for the Kingdom of Tonga, where parcel-level attribute extraction from satellite data is similarly the core task.

What satellite data cannot do here

SAR flood mapping struggles under dense forest canopy, where the double-bounce signal from tree trunks can mimic or mask open water. Wind roughening of flood surfaces produces backscatter that resembles dry land. Urban areas present the opposite problem: building facades create strong double-bounce returns that can obscure inundated streets at 10 m resolution.

Return-period modelling is only as good as the discharge frequency analysis feeding it. In data-sparse regions with short gauge records, the 1-in-200-year discharge estimate carries wide uncertainty intervals that propagate directly into the inundation extent. Satellite data cannot manufacture hydrological records that do not exist. And no remote sensing method currently resolves internal drainage features, culverts or flood defence structures at the scale that matters for individual parcels. Ground survey remains necessary for any parcel where the modelled exposure is close to a decision threshold.

Typical figures

Flood-extent spatial resolution	10 m (Sentinel-1 IW mode); 30 m for Landsat-derived water masks
Sentinel-1 revisit at mid-latitudes	1 to 3 days (two-satellite constellation); 6 days at equator
Terrain model vertical accuracy (GLO-30)	~4 m RMSE globally; 6 m or worse in flat coastal terrain
Terrain model horizontal resolution	30 m (Copernicus DEM GLO-30)
Flood-event detection latency	6 to 48 hours from satellite overpass to open-water mask
Sentinel-1 archive depth	From 2014 (Sentinel-1A launch); growing multi-event historical record
Minimum detectable inundated area	Approximately 0.5 ha at 10 m resolution under open-terrain conditions
Return periods modelled	1-in-50, 1-in-100, 1-in-200 year (subject to available discharge data)
Delivery formats	GeoTIFF flood masks, GeoPackage or Shapefile parcel-level risk attributes, CSV joined to asset register

Analytics Satellize can run

Historical flood-extent archive per geography	Log-ratio SAR backscatter change detection on Sentinel-1 IW scenes, morphological filtering	GeoTIFF stack of observed flood extents with event dates, covering the Sentinel-1 archive from 2014
Parcel-level flood-zone classification	Spatial intersection of modelled 1-in-100 and 1-in-200 year inundation extents with client asset register	GIS layer and CSV with probability band, inundated area (m²), and depth estimate per parcel
Hydraulic return-period inundation extents	1D/2D hydrodynamic modelling (HEC-RAS or LISFLOOD-FP class) on Copernicus DEM GLO-30, calibrated to observed SAR extents	Shapefile polygons for each return period with water surface elevation raster
DEM uncertainty flag per parcel	Terrain slope and GLO-30 error budget analysis; identification of parcels where horizontal boundary shift exceeds 50 m	Per-parcel flag column in asset register output; recommended LiDAR verification list
Expected annual damage estimate	Depth-damage curve application (published Environment Agency or JRC functions) to modelled inundation depth per parcel	Monetary EAD figure per parcel, suitable for input to discounted cash flow or loan-to-value adjustment
Near-real-time flood event alert	Automated SAR change detection triggered on new Sentinel-1 acquisitions over defined portfolio bounding boxes	Email or API alert with preliminary flood mask within 12 hours of overpass, flagging affected parcel IDs

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.