Floodplain inundation risk mapping for ground-based telecoms infrastructure
SAR-derived inundation history and LiDAR terrain models combine to score exchange buildings, base stations and cable ducts against return-period flood depths, with honest uncertainty where vertical accuracy is the binding constraint.
Sensors
- Sentinel-1 (ESA): C-band SAR at 5.405 GHz; Interferometric Wide Swath mode delivers 10 m ground range resolution across a 250 km swath. Dual-polarisation (VV+VH) backscatter distinguishes open water from vegetated or urban surfaces. Six-day repeat at mid-latitudes; cloud-independent acquisition makes it the primary inundation archive source, with imagery available back to 2014.
- TanDEM-X WorldDEM (DLR/Airbus): X-band bistatic InSAR-derived global DSM delivered as a bare-earth DTM (WorldDEM DTM product). Absolute vertical accuracy quoted at better than 2 m LE90 globally; relative accuracy better than 1 m LE90 over flat terrain. Grid spacing 12 m (standard) or 6 m (enhanced). The binding vertical constraint for sub-metre flood-depth estimation.
- ALOS-2 PALSAR-2 (JAXA): L-band SAR at 1.2 GHz. Longer wavelength penetrates partial vegetation canopy, improving inundation detection under riparian tree cover where C-band signal saturates. ScanSAR mode covers 350 km at 60 m resolution; Stripmap mode reaches 3 m in spotlight. Complements Sentinel-1 where vegetated floodplains obscure the inundation boundary.
- Copernicus Emergency Management Service (CEMS) / JRC Global Surface Water: The JRC Global Surface Water Explorer, derived from Landsat imagery 1984 to present, maps monthly water occurrence and seasonality at 30 m. Provides long-period flood recurrence statistics independent of the SAR archive, useful for return-period calibration where Sentinel-1 temporal coverage is insufficient for rare events.
- National airborne LiDAR surveys: Where available (Environment Agency in England and Wales, USGS 3DEP in the US, and equivalents), first- and last-return point clouds yield bare-earth DTMs at 0.25 m to 1 m posting with vertical accuracy of 0.05 to 0.15 m RMSE, far exceeding TanDEM-X. Coverage is national in some jurisdictions and absent in many others; this is the first question to resolve before any project begins.
Why telecoms infrastructure and floodplains keep finding each other
Telephone exchanges, cable head-ends and mobile base stations were sited for population coverage, not flood avoidance. Valleys concentrate people; rivers concentrate valleys. The result is a systematic overlap between the assets that carry the most traffic and the terrain that floods first. A single inundated exchange can silence tens of thousands of lines. A submerged cable duct carrying inter-city fibre is a multi-week repair at a point where access is, by definition, difficult.
The engineering question is not whether a site is in a floodplain but at what return-period depth it floods, how long it stays inundated, and whether the critical threshold is 0.1 m of water in a duct trench or 0.5 m against an exchange door. Those are answerable questions. SAR-based inundation mapping and high-resolution terrain modelling now make them answerable at scale, without waiting for the next flood event to reveal the answer the hard way.
What SAR backscatter actually detects, and where it stops
Open water returns very low C-band backscatter in calm conditions because the specular surface reflects energy away from the sensor. That contrast against rougher land surfaces is the detection mechanism. Sentinel-1 VV polarisation is most sensitive to surface roughness; VH adds sensitivity to volume scattering, helping separate shallow inundation under grass from dry ground. Threshold-based change detection against a pre-flood reference image is the standard approach, with the flooded class typically defined by a drop of 3 to 5 dB from baseline.
The limits are real and worth stating plainly. Urban areas produce strong double-bounce returns that can mask inundation beneath buildings. Dense riparian vegetation attenuates the signal; L-band PALSAR-2 partially compensates but does not eliminate the problem. Wind-roughened water surfaces can resemble dry ground. And SAR detects the water surface, not the depth: converting extent to depth requires the terrain model, which introduces its own uncertainty. In flat agricultural floodplains, 10 m SAR pixels resolve inundation boundaries adequately. In urban areas with narrow streets and varied building heights, 10 m is coarse and the boundary uncertainty widens.
Building the depth estimate: terrain accuracy is everything
Flood depth at any asset location is the difference between the water surface elevation and the bare-earth ground elevation at that point. The water surface elevation comes from hydraulic modelling or, in post-event analysis, from the SAR-derived inundation boundary intersected with the DTM (the so-called water-level fitting method). The ground elevation comes from the DTM. Both carry error, and the errors compound.
TanDEM-X WorldDEM achieves better than 2 m absolute vertical accuracy at LE90, meaning that a depth estimate of 0.8 m carries uncertainty that could plausibly encompass zero. This is not a flaw in the product; it is physics. The 12 m posting also means that a cable duct in a narrow street depression may not be resolved. Where national airborne LiDAR exists at 0.5 m posting and 0.1 m vertical RMSE, depth estimates become genuinely useful at the sub-metre level. The first step in any project scoping is therefore to determine which DTM is available for the territory in question, because the answer determines whether depth scoring is meaningful or merely indicative.
For return-period analysis, the SAR archive from Sentinel-1 (2014 to present) covers roughly a decade. That is sufficient to characterise frequent events (2-year to 10-year return periods) but statistically thin for 100-year or 200-year design events. JRC Global Surface Water, extending back to 1984, and national gauge records are used to extend the statistical base, but extrapolation to rare return periods always carries wider confidence intervals than the point estimates suggest.
Scoring assets: from probability surface to infrastructure risk register
The analytic output is a risk score per asset, not a binary flood/no-flood flag. Each exchange building, base station and cable duct segment is assigned an estimated inundation depth at the 10-year, 30-year, 100-year and 200-year return periods, derived from the ensemble of SAR-mapped flood extents and the hydraulic-terrain model. Depth is then compared against asset-specific damage thresholds: a ground-floor exchange may tolerate 0.05 m before water reaches critical equipment; a buried duct may be more vulnerable to prolonged saturation than to peak depth.
Assets are ranked by expected annual damage, a product of inundation probability, depth-damage function and asset replacement cost. This ranking drives capital prioritisation: which sites warrant flood barriers, which require equipment elevation, which need alternative routing. The GIS deliverable is a scored asset layer that can be loaded directly into a network management platform or a corporate GIS.
Honest limits and the questions worth asking before commissioning
Several failure modes are worth naming before any procurement decision. SAR-based inundation mapping performs poorly in urban canyons where building shadowing and layover corrupt the backscatter signal. Depth estimates in areas with only TanDEM-X coverage are indicative, not engineering-grade. The Sentinel-1 archive begins in 2014, so events before that date are not directly captured; historical flood outlines from national agencies should be incorporated where they exist.
Copernicus Emergency Management Service activations provide rapid flood mapping for declared disasters, but the turnaround is event-driven and the mapping extent is determined by the activation request, not by infrastructure asset distribution. For systematic pre-event risk scoring, a purpose-built inundation archive analysis is more appropriate than relying on CEMS outputs alone.
Satellize runs this analysis on open Sentinel-1 and JRC data, supplemented by commercial TanDEM-X WorldDEM or client-supplied LiDAR where available. The approach is the same one applied in the Tonga crop-estimation programme: open constellations do the heavy lifting; commercial data fills the gaps where the physics demands it.
What the output looks like in practice
Delivery is a scored GIS layer in GeoPackage or GeoJSON format, with accompanying depth-uncertainty rasters for each return period. Each asset record carries the estimated inundation depth, the depth uncertainty range, the return period at which inundation first occurs, and a traffic-light risk band. A summary report maps the highest-risk clusters and identifies whether the dominant exposure is exchange buildings, active base stations or passive duct infrastructure.
Update frequency depends on the use case. A one-time risk register for capital planning can be produced from the existing Sentinel-1 archive in weeks. A monitored service that flags new inundation events against the asset register within 24 to 48 hours of a Sentinel-1 acquisition is also feasible, given the six-day revisit and near-real-time data availability through the Copernicus Data Space Ecosystem.
Typical figures
| Primary SAR spatial resolution | 10 m (Sentinel-1 IW mode); 3 m available from PALSAR-2 Stripmap |
| SAR archive depth | Sentinel-1: 2014 to present; PALSAR-2: 2014 to present; JRC Global Surface Water (Landsat): 1984 to present |
| Sentinel-1 revisit at mid-latitudes | 6 days (single satellite); 3 days where Sentinel-1A and 1B both operational |
| DTM vertical accuracy (TanDEM-X WorldDEM) | Better than 2 m absolute LE90; better than 1 m relative LE90 over flat terrain; 12 m or 6 m posting |
| DTM vertical accuracy (airborne LiDAR, where available) | 0.05 to 0.15 m RMSE; 0.25 to 1 m point spacing depending on survey specification |
| Minimum detectable inundation depth | Not directly detectable by SAR; depth is modelled from extent plus DTM. Sub-metre depth estimates require LiDAR-grade terrain. |
| Return-period range | 2-year to 200-year; statistical confidence degrades significantly beyond 50-year where SAR archive is the sole input |
| SAR frequency bands used | C-band 5.405 GHz (Sentinel-1); X-band (TanDEM-X); L-band 1.2 GHz (PALSAR-2) |
| Delivery formats | GeoPackage, GeoJSON, GeoTIFF depth rasters, CSV asset risk register, PDF summary report |
| Event detection latency (monitored service) | 24 to 48 hours after Sentinel-1 acquisition, subject to data pipeline configuration |
Analytics Satellize can run
| SAR inundation frequency map | Threshold-based change detection on Sentinel-1 VV/VH backscatter time series against pre-flood reference; open-water pixels classified by dB drop from baseline | GeoTIFF raster of inundation frequency (fraction of acquisitions showing water) per 10 m pixel, covering the full Sentinel-1 archive |
| Return-period flood extent polygons | Empirical frequency analysis on SAR-derived inundation extents, supplemented by JRC Global Surface Water occurrence statistics and national gauge records where available | Vector polygon layers for 10-, 30-, 100- and 200-year return-period extents, with confidence bands |
| Flood depth probability rasters | Water-level fitting: SAR inundation boundary intersected with TanDEM-X or LiDAR DTM to derive water surface elevation; depth computed as difference from bare-earth terrain | GeoTIFF depth rasters per return period, with accompanying uncertainty rasters expressing LE90 depth range |
| Asset risk score layer | Point-in-polygon depth extraction for each asset location; depth compared against asset-class damage thresholds; expected annual damage computed from depth-damage function and inundation probability | GeoPackage asset layer with per-site depth estimates, uncertainty ranges, first-inundation return period and traffic-light risk band |
| Vegetated floodplain inundation supplement | ALOS-2 PALSAR-2 L-band backscatter analysis for riparian zones where C-band Sentinel-1 signal is attenuated by canopy; merged with Sentinel-1 extent to close detection gaps | Revised inundation extent polygons for vegetated floodplain segments, flagged where L-band data was required |
| Near-real-time flood event alert | Automated Sentinel-1 acquisition monitoring via Copernicus Data Space Ecosystem; change detection run against baseline within 24 to 48 hours of new acquisition; affected assets identified by spatial join | Alert feed (JSON or email) listing assets within the detected inundation extent, with estimated depth range and asset criticality rank |
| Capital prioritisation report | Ranking of assets by expected annual damage; cluster analysis to identify geographic concentrations of high-risk sites; comparison of mitigation options (barrier, equipment elevation, route diversity) against risk reduction | PDF report with ranked asset table, risk cluster maps and indicative mitigation priority list |
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.