Coastal erosion rate mapping for terrestrial cable route vulnerability assessment
Multi-decadal satellite waterline extraction and very-high-resolution photogrammetry quantify how fast a coast is retreating, identifying terrestrial cable sections at risk of exposure or severance before a fault occurs.
Sensors
- Landsat 8/9 (USGS/NASA): 30 m multispectral resolution, 16-day repeat at the equator. Archive extends to 1972 across the full Landsat series, enabling decadal shoreline change rates. Band 5 (NIR) provides a clear land/water boundary for waterline extraction, though 30 m pixels set a hard positional uncertainty floor of roughly 15 m per extracted shoreline position.
- Sentinel-2 (ESA): 10 m resolution in visible and NIR bands, 5-day revisit at mid-latitudes with both Sentinel-2A and 2B operating. Better suited than Landsat to resolving narrow beaches and barrier spits. Tidal normalisation is still required; the 10 m pixel sets positional uncertainty of 10 to 20 m per waterline extraction.
- Pleiades Neo (Airbus): 30 cm native resolution, daily revisit capability over tasked sites. Used for Structure-from-Motion photogrammetry when stereo or tri-stereo acquisition is commissioned, producing cliff-face digital elevation models accurate to 0.1 to 0.5 m vertically under good geometry. Cloud cover remains the primary operational constraint.
- WorldView-3 (Maxar): 31 cm panchromatic, 1.24 m multispectral, eight SWIR bands. SWIR bands can discriminate saturated sand from dry sand and help delineate the recent storm-swash zone. Stereo pairs support photogrammetric cliff-top recession mapping at sub-metre accuracy.
Why erosion rate matters more than erosion position
A cable buried 1.2 m below a cliff top is not in danger today. It may be in danger in four years. The question that matters for network resilience planning is not where the cliff edge currently sits but how fast it is retreating, and whether that rate is accelerating. A single satellite image answers neither question. A 30-year Landsat time series, properly tidal-corrected and georeferenced, can answer both.
Erosion physics differ sharply by coastal type, and the distinction matters for how you interpret satellite data. Sandy barrier coasts respond to individual storm events and can recover between seasons; a single post-storm image will overstate long-run retreat. Soft-cliff coasts, composed of glacial till, London Clay or similar materials, retreat episodically through rotational failures and face slumps, producing non-linear time series with quiet periods interrupted by sudden large losses. Hard-rock coasts retreat far more slowly, sometimes centimetres per decade, but when they fail they do so catastrophically. Each type demands a different sampling strategy and a different statistical treatment of the extracted waterline or cliff-top position series.
Extracting a shoreline from a pixel: what the numbers actually mean
Waterline extraction from multispectral imagery uses the contrast between water-absorbing NIR reflectance and land reflectance to locate the instantaneous water's edge. Applied to Landsat 8 at 30 m resolution, the extracted line carries a positional uncertainty of one to two pixels, meaning 15 to 60 m depending on the extraction algorithm and image quality. Sentinel-2 at 10 m reduces that range to roughly 10 to 20 m. These are not small numbers on a beach that retreats 0.5 m per year; a single image contributes more noise than signal. The value comes from stacking many images over many years and fitting a linear or piecewise trend through the position time series, which suppresses random error and reveals the underlying rate.
Tidal stage at the moment of image acquisition is the largest systematic bias. A shoreline extracted at mean high water will appear landward of one extracted at mean low water by a distance equal to the beach slope times the tidal range, which can easily exceed 50 m on a macrotidal coast. Correcting for this requires pairing each image timestamp with a tidal prediction from a validated hydrodynamic model such as the FES2014 global tide solution, then projecting the waterline to a common datum. Skipping this step produces spurious trends that can dwarf real erosion signals.
Cliff-top recession: where photogrammetry earns its place
For soft and hard cliff coasts, the relevant threat to a buried cable is not the waterline but the cliff-top edge. A cable routed 10 m inland of a cliff that retreats 2 m per year has a five-year exposure horizon. Mapping cliff-top position from medium-resolution imagery is unreliable because the edge is often a single-pixel feature obscured by vegetation overhang or shadow. Very-high-resolution stereo imagery from Pleiades Neo or WorldView-3 changes the analysis.
Stereo pairs separated by a suitable base-to-height ratio allow photogrammetric reconstruction of the cliff face as a dense point cloud or digital surface model. Differencing two such models acquired one or more years apart produces a volumetric change map. Published studies using similar workflows on UK and French chalk and clay cliffs have resolved individual failure events of a few hundred cubic metres. The practical limits are cloud cover at tasking time, the need for at least two acquisition epochs separated by a meaningful interval, and the fact that overhanging vegetation can mask the true cliff edge in both epochs.
Translating retreat rates into cable risk categories
Raw retreat rates in metres per year are not directly actionable for a cable operator. The useful output is a risk horizon: given the current burial depth and setback distance, and given the measured retreat rate with its confidence interval, how many years before the cable section is likely to be exposed? That calculation requires knowing the original cable route geometry, which may come from as-built survey records, and combining it with the satellite-derived erosion rate and its uncertainty bounds.
A typical output classifies cable sections into time-to-exposure bands. Sections with a median exposure horizon under ten years, given the measured rate, warrant physical inspection and possible rerouting. Sections with horizons of ten to thirty years require monitoring at regular intervals. Beyond thirty years, the uncertainty in projected rates dominates and the appropriate response is periodic reassessment rather than immediate action. The honest caveat is that episodic events, a single winter storm, a cliff failure triggered by groundwater, can compress a thirty-year horizon to zero in one night. Satellite-derived rates describe the background trend; they do not predict discrete failure events.
Practical limits and the cases where satellite data is not enough
Cloud cover is the most persistent operational constraint. Optical sensors cannot see through it, and the UK, Ireland and much of the North Atlantic coast, where many high-value terrestrial cable routes run, can go weeks without a usable acquisition. This is partly mitigated by the depth of the Landsat and Sentinel-2 archives: even a cloudy coast accumulates enough clear observations over years to support trend analysis. It is not mitigated for near-real-time monitoring of a specific storm event.
Synthetic aperture radar can detect the land/water boundary through cloud, but the shoreline extraction accuracy from C-band SAR such as Sentinel-1 is generally worse than from optical data on beaches with low backscatter contrast, and it contributes little to cliff-top mapping. For hard-rock coasts retreating at less than 0.1 m per year, even a 30-year Landsat archive may not resolve the trend above the positional noise floor; ground-based or airborne lidar is the appropriate tool in those cases. Satellize integrates satellite-derived rates with client-supplied ground-truth data and as-built cable records, as it does when combining satellite imagery with local agronomic knowledge in the Tonga crop-estimation programme.
From archive to deliverable: what the workflow produces
A typical engagement begins with archive harvesting: pulling all cloud-free or low-cloud Landsat and Sentinel-2 scenes covering the cable corridor for the full available record, currently back to 1984 for Landsat 5 and to 2015 for Sentinel-2. Each scene is atmospherically corrected, tidal-stage-tagged and waterline-extracted. The resulting position time series, one value per scene per transect, is then regressed to produce a rate and confidence interval at each transect, typically spaced at 50 m intervals along the coast.
Where the cable route passes soft or hard cliffs, very-high-resolution stereo tasking adds cliff-top recession data at sub-metre resolution. The two data streams are merged into a single GIS layer that attributes each cable segment with its erosion rate, uncertainty, coastal type classification and estimated time-to-exposure band. The final deliverable is that attributed GIS layer, a PDF technical report explaining the methodology and its limits, and, where the client requests it, a monitoring schedule specifying which segments to reassess and at what interval.
Typical figures
| Waterline extraction resolution (Landsat 8/9) | 30 m pixel; positional uncertainty 15 to 60 m per extracted shoreline |
| Waterline extraction resolution (Sentinel-2) | 10 m pixel; positional uncertainty 10 to 20 m per extracted shoreline |
| Cliff-top mapping resolution (Pleiades Neo / WorldView-3 stereo) | 30 to 50 cm native; DSM vertical accuracy 0.1 to 0.5 m under good stereo geometry |
| Revisit for trend analysis | Landsat 16 days; Sentinel-2 5 days (two satellites); VHR commercial on-demand |
| Archive depth | Landsat series from 1972 (usable multispectral from 1984); Sentinel-2 from 2015 |
| Minimum detectable retreat rate | Approximately 0.3 to 0.5 m/year from Sentinel-2 over a 10-year record; coarser from Landsat alone |
| Tidal normalisation requirement | Required for all multi-temporal comparisons; uses validated global tide model (e.g. FES2014) |
| Transect spacing (standard output) | 50 m along-shore; adjustable to 10 m for high-priority cliff sections |
| Delivery format | GeoPackage or Shapefile with attributed erosion rates; PDF technical report; optional GeoTIFF DSM difference rasters |
| Cloud cover constraint | Optical sensors blocked by cloud; mitigated by multi-year archive depth, not by near-real-time tasking |
Analytics Satellize can run
| Decadal shoreline change rate map | Waterline extraction from Landsat and Sentinel-2 NIR band, tidal normalisation against FES2014, linear regression per transect | GIS polyline layer with retreat rate (m/year) and 95% confidence interval attributed to each 50 m transect |
| Cliff-top recession model | Structure-from-Motion photogrammetry from Pleiades Neo or WorldView-3 stereo pairs, multi-epoch DSM differencing | GeoTIFF volumetric change raster and cliff-edge recession vector for each acquisition epoch pair |
| Coastal type classification | Supervised classification of spectral and textural features from Sentinel-2 and VHR imagery into sandy barrier, soft-cliff and hard-rock classes | GIS polygon layer with coastal type and associated erosion-physics flag, used to weight risk calculations |
| Cable segment risk horizon assessment | Geometric intersection of cable route as-built positions with erosion rate raster; time-to-exposure calculation using burial depth and setback distance from client records | Attributed cable route GIS layer with time-to-exposure band (under 10 years / 10 to 30 years / over 30 years) and recommended action flag |
| Storm-event shoreline displacement detection | Pre- and post-storm waterline extraction from available cloud-free Sentinel-2 acquisitions; comparison against long-run trend baseline | Event report quantifying acute displacement relative to background retreat rate, delivered within days of cloud-clear acquisition |
| Monitoring schedule and reassessment trigger | Statistical analysis of rate uncertainty and confidence interval width to determine minimum interval at which new data meaningfully updates the trend estimate | Written monitoring protocol specifying reassessment dates and cloud-free acquisition targets per priority cable segment |
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.