Coastal cliff and shoreline erosion rate measurement
Soft-rock cliffs can lose metres in a single storm. Repeat satellite SAR, very-high-resolution optical imagery, and photogrammetric change detection now make sub-annual recession budgets tractable without a single site visit.
Sensors
- Sentinel-1 SAR (C-band, ESA): 6-day repeat at mid-latitudes in interferometric wide-swath mode, 5 x 20 m ground range resolution. Coherence differencing between acquisitions flags surface disturbance from cliff-top collapse; backscatter change detects beach sediment redistribution. Cannot resolve sub-metre planimetric shift directly, but coherence loss is a reliable collapse indicator.
- Sentinel-2 MSI (ESA): 10 m visible and NIR bands, 5-day revisit with combined Sentinel-2A/B. Adequate for shoreline planimetric mapping where recession exceeds roughly 10 m per measurement interval, and for tracking turbidity plumes after mass failures. Cloud cover over UK coasts routinely reduces usable acquisition frequency to one clear pass per two to four weeks.
- Planet SuperDove: 3 m resolution, daily revisit, eight spectral bands including coastal blue and red-edge. Sufficient to detect cliff-top edge position changes of 3 to 5 m and to map beach planimetric area with reasonable confidence. Radiometric consistency across the constellation has improved markedly since the SuperDove generation, supporting multi-year time series.
- WorldView-3 (Maxar): 0.31 m panchromatic, 1.24 m multispectral, tasked on demand. At this resolution, stereo pairs yield photogrammetric surface models accurate to roughly 0.3 to 0.5 m vertical RMS over bare cliff faces, enabling volume-change estimates that approach the precision of airborne lidar for annual budget calculations. Revisit is opportunity-dependent, not guaranteed.
Why soft-rock cliffs defeat conventional monitoring
Chalk and glacial till cliffs do not erode continuously. They accumulate stress, then fail catastrophically, sometimes losing three to five metres of cliff top in a single winter storm. That episodic character means annual average recession rates, which run at roughly 0.1 to 1.5 m per year along parts of the Yorkshire and Norfolk coasts according to Environment Agency long-term records, conceal enormous event-to-event variance. A monitoring scheme that visits a site twice a year will almost certainly miss the failure itself and see only the aftermath.
Satellite constellations with daily or near-daily revisit change that calculus. The question shifts from 'how much did it erode this year?' to 'which storm caused which failure, and how large was the detachment?' That distinction matters for coastal engineers designing setback lines and for insurers pricing properties within the recession envelope.
What a floating roof gives away: SAR coherence as a collapse detector
Synthetic aperture radar coherence measures how similar the radar phase is between two passes over the same ground. Stable bare rock and compacted soil maintain high coherence over six-day Sentinel-1 intervals. A cliff-face collapse, or the slumping of saturated till onto a beach, randomises the phase completely. The coherence drops to near zero over the disturbed area, and the spatial extent of that drop is a proxy for the failure footprint.
The limit is resolution. Sentinel-1's 5 x 20 m pixel cannot resolve a failure narrower than roughly 20 m in azimuth. Small topples that are geomorphologically significant but spatially compact will be missed. SAR coherence is therefore most useful as a coarse screening tool: it flags that something happened, and when, with six-day precision. Optical imagery then provides the geometry.
Planimetric shoreline mapping: where resolution actually matters
Extracting a cliff-top edge from satellite imagery is a classification problem. The edge is the boundary between vegetated or cultivated land and bare cliff face or open air. At 10 m resolution (Sentinel-2), that boundary can be located to roughly plus or minus one pixel, meaning positional uncertainty of up to 10 m. For a site retreating at 0.5 m per year, a decade of Sentinel-2 data is the minimum needed to produce a statistically meaningful trend, and only if cloud-free acquisitions are distributed evenly through the period.
Planet SuperDove at 3 m reduces positional uncertainty to roughly 3 to 5 m, making annual recession detectable at moderately eroding sites. WorldView-3 pushes this further: sub-metre planimetric mapping becomes feasible, and stereo-derived digital surface models allow cliff-face volume to be estimated directly rather than inferred from planimetric retreat alone. The trade-off is cost and tasking latency. WorldView-3 is not a monitoring constellation; it is a precision measurement instrument deployed on demand.
A practical programme combines the two. Sentinel-2 and SuperDove provide the continuous time series that catches the timing of events. WorldView-3 stereo is commissioned immediately after a significant failure to capture the fresh scar geometry before wave action and weathering obscure it.
Photogrammetric volume budgets and the lidar baseline question
The Environment Agency's national lidar programme has produced multiple-epoch airborne surveys of most of the English coastline at point densities of 1 to 4 points per square metre, with vertical accuracy around 0.1 m over flat ground. These surveys are the gold standard for cliff-volume change measurement. Satellite photogrammetry from WorldView-3 stereo pairs approaches but does not match that accuracy over complex cliff faces, where occlusion and shadow degrade the point cloud.
The practical value of satellite photogrammetry is not to replace airborne lidar but to fill the gaps between surveys. Airborne campaigns over UK coastal sites typically occur every three to five years. A WorldView-3 stereo acquisition can be scheduled within days of a reported failure. The resulting surface model, differenced against the most recent lidar baseline, gives a failure volume estimate accurate enough for engineering purposes even if it is not publication-quality geomorphology.
One honest caveat: cliff faces that overhang, or that have complex re-entrant geometry, will have systematic gaps in any photogrammetric model derived from near-nadir satellite geometry. Airborne oblique imagery or structure-from-motion from UAVs remains superior for those specific situations.
Building a recession time series that holds up to scrutiny
Shoreline change analysis from multi-sensor time series carries several sources of error that are easy to understate. Tidal stage at image acquisition changes the apparent shoreline position at the beach toe by several metres on a macrotidal coast. Seasonal vegetation growth advances or retreats the detectable cliff-top edge by one to two metres independently of actual erosion. Atmospheric correction differences between sensors introduce apparent reflectance shifts that affect edge-detection algorithms.
Rigorous programmes correct for tidal stage using co-located tide gauge records or modelled tidal surfaces, mask acquisitions outside a defined tidal window, and apply consistent edge-detection algorithms across the full archive. The USGS CoastSat toolkit, published in peer-reviewed literature and openly available, implements much of this workflow for Landsat and Sentinel-2. Extending it to Planet and WorldView requires additional radiometric normalisation steps but the underlying method is transferable.
Satellize applies this class of multi-sensor time-series analysis, including the tidal and seasonal corrections described above, as part of its geohazard analytics work. The Tonga crop-estimation programme sits in a different domain, but the same discipline of handling multi-sensor radiometric consistency applies across both.
Honest limits and what they mean for a procurement decision
No satellite system currently detects annual recession rates below roughly 0.3 m per year with statistical confidence at individual cliff segments, unless a very long archive (fifteen or more years) is available and cloud cover is low. For the UK, cloud is a persistent constraint. A site on the Yorkshire coast may yield only eight to twelve cloud-free Sentinel-2 acquisitions per year, and fewer still in winter when storms drive the most erosion.
SAR is cloud-independent, which is its main attraction for UK coastal work, but its resolution floor means it complements rather than replaces optical methods. The combination of Sentinel-1 coherence monitoring for event detection, SuperDove for planimetric time series, and periodic WorldView-3 stereo for volume measurement is the most defensible architecture for a national or regional coastal erosion programme. Each component has a specific, non-overlapping function. Expecting any single sensor to do all three jobs will produce a programme that does none of them well.
Typical figures
| Planimetric resolution (Sentinel-2) | 10 m visible/NIR bands; cliff-top edge location uncertainty ±10 m |
| Planimetric resolution (Planet SuperDove) | 3 m; edge location uncertainty ±3–5 m |
| Planimetric resolution (WorldView-3) | 0.31 m panchromatic; stereo DSM vertical accuracy ~0.3–0.5 m RMS over bare cliff |
| SAR coherence pixel (Sentinel-1 IW) | 5 m range × 20 m azimuth; minimum detectable failure footprint ~20 m |
| Revisit cadence | Sentinel-1: 6 days; Sentinel-2: 5 days (A+B combined); Planet: daily; WorldView-3: tasked on demand |
| Minimum detectable annual recession | ~0.3 m/yr with SuperDove over 3+ years; ~10 m/yr with Sentinel-2 over 1 year |
| Cloud impact (UK coastline) | Optical usable acquisition rate typically 8–12 cloud-free passes/year for Sentinel-2 at UK latitudes |
| Archive depth | Sentinel-1/2: from 2014/2015; Landsat: from 1972; Planet SuperDove: from ~2021; WorldView-3: from 2014 (tasked only) |
| Delivery formats | GeoTIFF change-detection layers, GeoJSON shoreline polylines, CSV recession-rate tables, PDF epoch-comparison reports |
| Tidal correction requirement | Mandatory for beach-toe shoreline; co-located tide gauge or modelled tidal surface needed; cliff-top edge less sensitive |
Analytics Satellize can run
| Cliff-top recession rate map | Multi-epoch planimetric edge detection on SuperDove or WorldView-3 time series, tidal and seasonal correction applied, linear regression per cliff segment | GeoJSON polyline layer with recession rate (m/yr) and confidence interval per 50 m segment; annual update or event-triggered |
| Failure event detection alert | Sentinel-1 coherence differencing between consecutive 6-day pairs; threshold exceedance flags disturbed area | Alert feed with failure location, approximate footprint area, and acquisition date; delivered within 24 hours of SAR pass processing |
| Post-failure volume estimate | WorldView-3 stereo photogrammetry differenced against Environment Agency lidar baseline; DSM subtraction for volume calculation | GeoTIFF elevation-difference raster and PDF summary with estimated failure volume (m³) and scar geometry |
| Shoreline planimetric change time series | CoastSat-class workflow applied to Sentinel-2 and Planet archive; tide-windowed acquisition selection, sub-pixel edge detection | CSV time series of shoreline position per transect, with uncertainty bounds; GeoTIFF epoch-pair difference images |
| Beach volume proxy from SAR backscatter | Sentinel-1 VV/VH backscatter change over intertidal zone correlated with sediment moisture and surface roughness proxies for qualitative sediment budget tracking | Monthly GeoTIFF backscatter-anomaly layer; qualitative sediment redistribution report |
| Multi-decadal recession trend (Landsat baseline) | Landsat 5/7/8/9 archive from 1984 onward processed through consistent NDWI-based shoreline extraction; long-term linear and non-linear trend fitting per transect | GeoJSON recession-rate layer spanning up to 40 years; PDF trend report with storm-event annotations where meteorological records permit |
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.