Coastal erosion monitoring for shoreline property risk
Satellite time-series can quantify shoreline retreat rates to sub-metre precision per year, giving property owners, insurers and planners a defensible, multi-decadal record of coastal land loss before a single survey boat is hired.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands; 5-day revisit at mid-latitudes with two satellites. The near-infrared band (Band 8, 842 nm) is the primary waterline extraction channel. Free archive from 2015.
- Landsat 8/9 OLI: 30 m multispectral resolution; 16-day revisit per satellite, 8-day combined. Near-infrared Band 5 used for waterline extraction. Archive extends to 1972 across the Landsat programme, enabling multi-decadal trend analysis unavailable from any other open source.
- Planet SuperDove: 3 m resolution; near-daily revisit globally. Useful for resolving fine-scale features such as narrow spits or groyned compartments where 10 m pixels average across the waterline. Commercial licence required; archive depth varies by site.
- Sentinel-1 SAR (C-band): IW mode delivers 10 m range resolution; 6-day revisit (two satellites). Cloud-independent, so it fills temporal gaps left by optical sensors in persistently overcast climates. Backscatter contrast at the water-land boundary enables waterline mapping regardless of illumination.
What the near-infrared band gives away
Water absorbs near-infrared radiation strongly. Dry or vegetated land reflects it. That contrast, exploited through the Normalised Difference Water Index (NDWI = (Green - NIR) / (Green + NIR)) or simple NIR thresholding, produces a binary land-water mask from any multispectral image. The waterline is the boundary of that mask.
At Sentinel-2's 10 m pixel size, a shoreline that crosses a pixel diagonally sits somewhere inside it, not neatly at its edge. Sub-pixel methods, particularly the USGS Digital Shoreline Analysis System (DSAS), address this by fitting a statistical transect across the gradient zone rather than snapping to a pixel boundary. Applied to a stack of Sentinel-2 images, DSAS can resolve net shoreline movement rates to roughly 0.5 m per year when the time-series is long enough and tidal correction is applied. With Landsat's 30 m pixels the rate uncertainty rises, but the archive depth back to the 1980s more than compensates for the coarser geometry when the goal is decadal trend detection.
The tidal problem, and why ignoring it invalidates a survey
Every satellite image is acquired at a specific moment in the tidal cycle. A beach with a 1:50 intertidal gradient and a 2 m tidal range will show a waterline that shifts 100 m horizontally between low and high tide. That variation dwarfs the annual erosion signal on most coasts. An uncorrected time-series will show apparent shoreline oscillation of tens of metres driven entirely by tidal stage, not land loss.
Correction requires two inputs: the predicted tidal height at image acquisition time (available from global tide models such as FES2014 or TPXO, accurate to roughly 10 cm RMS in open-ocean settings) and a local intertidal slope estimate derived from the image stack itself or from a digital elevation model. The corrected waterline is then reduced to a common datum, typically mean sea level or mean high water. This step is not optional. Any erosion rate presented without tidal correction should be treated as unreliable for property risk purposes.
Sentinel-1 SAR offers a partial workaround. Because it acquires at fixed local solar times, the tidal phase at acquisition is more predictable than for optical sensors, and the same tidal correction logic applies. Its real advantage is cloud penetration: in places like the UK's west coast or the Pacific Islands, optical sensors may yield fewer than 20 usable images per year, while SAR delivers a consistent fortnightly cadence regardless of cloud.
Building a multi-decadal rate from mixed sensors
A credible erosion rate for property risk needs at least 20 to 30 years of data. No single open satellite archive covers the full period at useful resolution. The practical approach is to anchor the long trend with Landsat (1984 to present at 30 m), then refine recent years with Sentinel-2 (2015 to present at 10 m) and, where budget allows, Planet SuperDove for the most recent 12 to 24 months.
Mixing sensors introduces inter-calibration error. Waterline positions extracted from Landsat 30 m pixels and Sentinel-2 10 m pixels will disagree by up to half a pixel even on the same date, simply due to geometric registration differences. The standard mitigation is to express all positions as transect-normal distances from a fixed baseline, then fit a weighted linear regression to the full time-series. The regression slope is the net shoreline movement rate in metres per year; the confidence interval quantifies the combined uncertainty from tidal correction residuals, pixel geometry and image-to-image co-registration error. Rates below roughly 0.3 m per year are generally not distinguishable from noise in a mixed Landsat-Sentinel-2 series.
Translating retreat rates into property exposure
A retreat rate in metres per year becomes a property risk metric when projected forward to a planning horizon and overlaid on cadastral geometry. A parcel boundary 15 m from the current cliff edge, eroding at 0.8 m per year, has roughly 18 years before the boundary is reached under a linear projection. Compound that with sea-level rise projections and the picture shortens further.
Linear projection is itself an honest approximation. Erosion is episodic: a single winter storm can remove in hours what took a decade to accumulate. The satellite time-series captures the net trend but not the episodic variance. For high-value properties, the satellite-derived rate should be treated as a baseline for probabilistic exposure modelling, not a deterministic forecast. Presenting it as the latter to a buyer or underwriter would be misleading.
Sentinel-1 coherence change detection adds a complementary signal. Sudden loss of SAR coherence between two acquisitions can flag cliff-face collapse or beach-crest overwash events at roughly 10 m spatial resolution, providing near-real-time alerts between the slower optical updates.
Honest limits of the method
Cloud cover is the dominant operational constraint for optical sensors. In high-latitude or tropical sites, months-long cloud gaps reduce the effective temporal resolution to seasonal at best. SAR fills some of this gap but requires its own correction pipeline and does not directly substitute for NIR-based waterline extraction.
Rocky coastlines with steep profiles are poorly served by intertidal slope correction because the tidal excursion is small relative to pixel size. The method works best on sandy and mixed sediment beaches with gentle gradients. Armoured coastlines, where seawalls or revetments fix the shoreline position, require a different analytical approach because the satellite-detected waterline reflects the structure, not natural erosion.
Archive depth varies by geography. Landsat coverage is global but early scenes (pre-1990) have patchy cloud-free availability in tropical regions. Sentinel-2 is consistent from 2017 onwards; earlier scenes exist but with lower revisit frequency. Planet's commercial archive typically runs from 2016 at useful resolution.
Satellize runs DSAS-based shoreline analysis on open Sentinel and Landsat archives, with optional Planet tasking for recent high-resolution epochs. The Tonga crop-estimation programme demonstrated the same multi-sensor compositing pipeline in a cloud-affected Pacific environment.
Typical figures
| Spatial resolution (optical) | 10 m (Sentinel-2), 30 m (Landsat 8/9), 3 m (Planet SuperDove) |
| Spatial resolution (SAR) | 10 m range x 10 m azimuth (Sentinel-1 IW mode) |
| Revisit cadence | 5 days (Sentinel-2, two satellites); 8 days (Landsat 8+9 combined); near-daily (Planet); 6 days (Sentinel-1, two satellites) |
| Archive depth | 1972 to present (Landsat programme); 2015 to present (Sentinel-2); 2014 to present (Sentinel-1) |
| Waterline detection limit | Net movement rates above ~0.3 m/yr distinguishable from noise in mixed Landsat-Sentinel-2 series; ~0.5 m/yr uncertainty typical for single-sensor Sentinel-2 DSAS |
| Tidal correction accuracy | FES2014/TPXO global tide models: ~10 cm RMS in open-ocean settings; higher error in complex estuarine or enclosed-sea geometries |
| Spectral bands used | NIR (Sentinel-2 Band 8 at 842 nm; Landsat Band 5 at 865 nm) for NDWI; Green band paired for waterline extraction |
| SAR frequency | C-band, 5.405 GHz (Sentinel-1) |
| Minimum analysable shoreline length | Typically 500 m or more for statistically meaningful transect regression; shorter segments have high per-transect uncertainty |
| Deliverable latency | Historical rate products: days to weeks depending on archive volume; SAR coherence alerts: 1 to 2 days after acquisition |
Analytics Satellize can run
| Multi-decadal shoreline retreat rate map | DSAS transect regression on tidal-corrected Landsat + Sentinel-2 waterline stack; net shoreline movement (NSM) and linear regression rate (LRR) per transect | GIS vector layer (GeoPackage or Shapefile) with per-transect NSM, LRR and 95% confidence intervals; PDF summary report |
| Tidal-corrected waterline time-series | NIR thresholding and NDWI applied to atmospherically corrected imagery; waterline reduced to mean sea level datum using FES2014 tide model and DEM-derived intertidal slope | Dated GeoJSON waterline archive per site; CSV of transect positions with tidal stage metadata |
| Property parcel exposure ranking | LRR rate projected to 10, 25 and 50-year horizons; spatial intersection with cadastral parcel boundaries | Ranked parcel table with projected encroachment dates and confidence bands; exportable to Excel or GIS |
| Cliff-collapse and overwash event alerts | Sentinel-1 coherence change detection between consecutive 6-day SAR pairs; anomaly flagged when coherence drops below site-calibrated threshold | Near-real-time alert (email or API) with event location, approximate extent and acquisition date |
| Seasonal and storm-driven shoreline variability envelope | Percentile analysis (10th, 50th, 90th) of tidal-corrected waterline positions across full time-series to separate episodic from trend signal | Polygon layer showing stable envelope vs. net loss zone; used to contextualise linear trend for underwriting purposes |
| High-resolution epoch comparison (Planet SuperDove) | Sub-pixel waterline extraction at 3 m from Planet imagery for most recent 12 to 24 months; differenced against Sentinel-2 baseline | Change raster and vector overlay showing recent acceleration or stabilisation relative to long-term trend |
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.