Flood inundation frequency mapping for renewable energy site risk screening
Standard elevation models cannot tell a financier how often a solar farm sits under water. Multi-year Sentinel-1 SAR time series map historical inundation extent and frequency at 10 m, cloud-free, and fused with TanDEM-X terrain data to estimate return-period depths.
Sensors
- Sentinel-1 A/B SAR (C-band, 5.405 GHz): Interferometric Wide Swath mode delivers 10 m × 10 m ground range detected imagery at 250 km swath width. Repeat pass every 6 days at the equator with both satellites operational, 12 days with one. Cloud-penetrating C-band backscatter drops sharply over open water, giving a reliable inundation signal regardless of weather or season.
- TanDEM-X Digital Elevation Model: Global DEM at 12 m posting (0.4 arcsec) with vertical accuracy better than 2 m relative height error over flat terrain, published by DLR. Used to convert inundation extent polygons into approximate water-surface elevations and to estimate flood depth at candidate site footprints.
- SRTM (Shuttle Radar Topography Mission): 30 m global DEM freely available via USGS EarthExplorer. Older and coarser than TanDEM-X but sufficient for catchment delineation and slope classification. Vertical accuracy is approximately plus or minus 16 m at 90 percent confidence globally, degrading in vegetated or complex terrain.
- Copernicus Emergency Management Service (CEMS): CEMS Rapid Mapping activations produce validated flood extent polygons for major events, typically within 24 to 48 hours of tasking. The CEMS Risk and Recovery component provides historical delineations for past events going back to 2012, usable as ground-truth layers for SAR classifier calibration.
Why a 100-year floodplain map is not enough
National floodplain designations are built from hydraulic models calibrated to gauge records that can be sparse, outdated, or simply absent in many emerging markets. They classify land as inside or outside a zone. They say almost nothing about how often a specific parcel floods, how deep the water gets, or how long it lingers. A solar farm developer who relies on that binary designation to satisfy a lender is carrying hidden risk.
Satellite SAR changes the question. Instead of asking whether a site is in a floodplain, you ask: across the last decade of radar passes, how many times did open-water backscatter appear at this location, in which months, and at what spatial extent? That is an empirical frequency map, not a modelled probability. It is also independent of whether a national authority ever mapped the area at all.
What C-band backscatter gives away about standing water
Dry bare soil and dry vegetation return a diffuse backscatter signal in the range of roughly minus 10 to minus 15 dB in VV polarisation for Sentinel-1 IW mode. Open calm water acts as a specular reflector: the radar pulse bounces away from the sensor, and the returned signal drops to minus 20 dB or below. That contrast, typically 5 to 10 dB, is the detection mechanism. It is physically reliable and does not depend on optical wavelengths, so cloud cover is irrelevant.
The method is not without limits. Dense emergent vegetation such as reeds or flooded maize canopies can double-bounce radar energy back to the sensor, masking the water beneath and producing a false dry-land reading. Shallow sheet flooding of less than roughly 5 to 10 cm depth may not depress backscatter enough to cross a detection threshold, particularly over rough or vegetated surfaces. Wind-roughened water surfaces can also reduce the backscatter contrast and produce missed detections. These are honest constraints that must be declared in any risk report.
Building a ten-year inundation frequency stack
Sentinel-1 data are available from late 2014 onwards via the Copernicus Data Space. For a candidate site, the standard workflow acquires all available IW Ground Range Detected scenes in VV and VH polarisation, applies radiometric calibration and terrain correction referenced to the TanDEM-X DEM, and then applies an automated threshold classifier, often supplemented by a change-detection step that flags pixels whose backscatter falls below a site-specific dry-state baseline. Each scene produces a binary inundation mask.
Stack those masks across several hundred acquisitions and you get an inundation frequency raster: each pixel carries a value from zero (never detected as flooded) to one (flooded in every available scene). Aggregated by calendar month, the stack reveals seasonality. Aggregated by year, it reveals whether flood frequency is trending. For a 100 MW solar farm footprint, the relevant output is the percentage of the site area that has been inundated at least once, at least five times, and at least ten times across the archive. Those three numbers, combined with the corresponding dates, are what an insurance underwriter or project finance team actually needs.
Fusing the inundation extent polygons with TanDEM-X elevation data allows a rough flood-depth estimate at each event: the water-surface elevation is interpolated from the perimeter of the detected flood, and depth is the difference between that surface and the terrain beneath each inundated pixel. The uncertainty on this depth estimate is dominated by DEM vertical error and by the assumption of a flat water surface, which breaks down in high-gradient channels. Treat depth outputs as order-of-magnitude guidance, not engineering-grade survey data.
From frequency map to return-period risk layer
A ten-year SAR archive covers roughly 180 to 200 Sentinel-1 passes per location in mid-latitudes. That is enough to characterise events down to roughly a one-in-three-year return period with reasonable confidence. Estimating a one-in-fifty or one-in-hundred-year return period from satellite data alone is not statistically credible; the archive is too short. The correct approach is to use the satellite-derived frequency map to validate and spatially extend a hydraulic model that is calibrated to gauged events, or to anchor the satellite results to CEMS historical event polygons where those exist.
The output that matters for project finance is a site-level risk matrix: for each return period that the lender specifies (commonly 1-in-10, 1-in-25, 1-in-100 years), what fraction of the site is inundated, to what estimated depth, and for how long historically did comparable events persist? Duration is recoverable from the SAR time series by counting consecutive flooded acquisitions and multiplying by the revisit interval.
Integrating the analysis into a siting decision
The flood frequency layer is one input among several in a site-screening workflow. It is most useful when overlaid on the planned array layout so that individual inverter stations, cable trenches, and substation pads can be checked against their specific inundation histories rather than the site centroid alone. A substation sited on a micro-topographic high that has never appeared in the SAR flood stack carries a materially different risk profile from one in a shallow depression that flooded in four of the last ten years.
Satellize runs this analysis as part of broader site-screening packages, applying the same SAR-stack methodology used in its Tonga crop-estimation programme to multi-temporal change detection. The deliverable is a GIS layer set accompanied by a structured risk summary keyed to the parameters lenders and insurers typically request. If the site has already been selected and construction is under way, the same archive can establish a pre-construction baseline against which post-construction drainage performance can be assessed.
What this analysis cannot do
SAR-derived flood mapping is not a substitute for a site-specific hydrological survey or an engineering-grade flood-risk assessment. It cannot detect subsurface waterlogging, soil saturation, or groundwater ingress that does not produce surface ponding. It will undercount floods that occurred entirely between satellite passes, which at 6 to 12 day revisit is a genuine gap for fast-rising flash floods that drain within 48 hours. In heavily vegetated floodplains, the vegetation masking problem described earlier can produce systematic underestimates of inundation frequency.
Used correctly, the method reduces uncertainty and narrows the scope of expensive ground surveys. It does not eliminate the need for them. A site that shows zero inundation in the SAR archive still warrants a drainage assessment; it simply enters that assessment with a stronger prior.
Typical figures
| Spatial resolution (SAR inundation mask) | 10 m × 10 m (Sentinel-1 IW GRD) |
| Temporal revisit (Sentinel-1) | 6 days with two satellites; 12 days with one |
| Archive depth | Sentinel-1A from late 2014; approximately 10 years of usable passes |
| Radar frequency / polarisation | C-band 5.405 GHz; VV and VH polarisation used |
| DEM vertical accuracy (TanDEM-X) | Better than 2 m relative height error on flat terrain (DLR specification) |
| DEM spatial resolution | 12 m posting (TanDEM-X); 30 m (SRTM) |
| Minimum detectable flood depth (indicative) | Approximately 5 to 10 cm on bare soil; higher under vegetation canopy |
| Swath width (Sentinel-1 IW) | 250 km; continental coverage in single acquisition |
| Cloud sensitivity | None; C-band SAR is unaffected by cloud or precipitation |
| Deliverable formats | GeoTIFF inundation frequency rasters, GeoPackage event polygons, structured PDF risk summary |
Analytics Satellize can run
| Inundation frequency raster | Threshold-based SAR backscatter classifier applied to multi-temporal Sentinel-1 VV stack; change-detection baseline normalisation | GeoTIFF at 10 m resolution, pixel values 0 to 1 representing fraction of archive scenes flagged as flooded |
| Seasonal inundation calendar | Monthly aggregation of binary flood masks across full archive, identifying peak inundation months and inter-annual variability | 12-band raster (one per calendar month) plus tabular summary of mean flooded area by month |
| Event-level flood extent polygons | Per-scene flood mask vectorised and attributed with acquisition date, estimated peak extent, and duration derived from consecutive flooded passes | GeoPackage with event attribute table; compatible with QGIS and ArcGIS Pro |
| Indicative flood depth grid | Water-surface elevation interpolated from inundation perimeter using TanDEM-X DEM; depth computed as water surface minus terrain elevation per pixel | GeoTIFF depth raster with uncertainty band annotation; caveats on DEM vertical error included in metadata |
| Site-level risk matrix | Zonal statistics of inundation frequency and depth layers clipped to proposed array layout polygons; return-period extrapolation anchored to CEMS historical events where available | Structured PDF report with tables keyed to lender-standard return periods (1-in-10, 1-in-25, 1-in-100 year) |
| Infrastructure-component flood exposure overlay | Point-in-polygon intersection of inverter, substation, and cable-trench locations against inundation frequency raster | Annotated GIS layer with per-component flood frequency scores; exportable to project design software |
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.