Coastal land reclamation volumetric change detection

Multispectral time-series and SAR intertidal analysis let analysts measure the rate, extent and approximate volume of artificial land creation from orbit, with archives stretching back to 1984 and boundary precision reaching sub-metre scale.

Sensors

Sentinel-2 (ESA): 10 m resolution in visible and near-infrared bands; 5-day revisit at the equator with both satellites. The near-infrared band (Band 8) gives sharp water-land contrast for automated shoreline extraction. Free archive from 2015.
Landsat 8 and 9 (USGS/NASA): 30 m multispectral resolution; 8-day combined revisit. The archive extends to 1984 (Landsat 5), making it the only freely available source for decadal baselines on reclamation projects that began before the Sentinel era.
Sentinel-1 (ESA C-band SAR): 6-day repeat in Interferometric Wide Swath mode at 10 m resolution. Cloud-independent backscatter and coherence distinguish intertidal sediment from open water and from consolidated fill, which Landsat and Sentinel-2 cannot do reliably under cloud cover.
Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral. Resolves individual revetment blocks, cofferdam walls and fill boundaries. Stereo pairs allow photogrammetric surface models, though these capture only the above-water portion of fill.
Planet SuperDove: 3–4 m resolution, daily revisit. Useful for tracking rapid fill progression week by week and for cloud-gap-filling in tropical regions where Sentinel-2 coverage is frequently obscured.

What the water-land boundary actually records

Shoreline extraction from multispectral imagery rests on a simple physical fact: water absorbs near-infrared radiation almost completely while dry land reflects it. The Normalised Difference Water Index (NDWI), computed from green and near-infrared bands, produces a binary water mask from a single Sentinel-2 scene in minutes. Apply that mask to every cloud-free scene in a time-series and you have a dated record of every pixel that transitioned from water to land.

The catch is tidal stage. A pixel classified as land at low tide may be intertidal mudflat, not reclaimed ground. Reliable reclamation mapping therefore requires either tidal-stage filtering, restricting analysis to scenes acquired within a known tidal window, or explicit intertidal modelling using the full range of water masks across a tidal cycle. The latter approach, demonstrated at scale in published work using Landsat archives, can map intertidal zones to roughly 30 m horizontal accuracy, which is adequate for national-scale monitoring but not for boundary disputes measured in tens of metres.

SAR fills the gaps cloud leaves open

Tropical reclamation sites, which include many of the most contested ones, sit under persistent cloud for months at a time. Sentinel-1 C-band SAR penetrates cloud and acquires coherent imagery regardless of illumination. Consolidated fill, particularly compacted sand or aggregate, produces strong, stable backscatter. Open water produces near-zero backscatter in calm conditions. Intertidal sediment sits between the two and shifts with each pass.

Coherence change detection adds a further dimension. A newly placed fill surface has low coherence between successive Sentinel-1 passes because the surface is being actively worked. Once fill consolidates, coherence rises. That temporal coherence signature can distinguish active reclamation from completed reclamation, which is useful when the legal question is not just how much land exists but when it was created.

SAR does not resolve fine boundary geometry as well as optical imagery. At 10 m pixel spacing, a revetment wall a few metres wide is sub-pixel. For boundary precision at that scale, tasked very-high-resolution optical imagery is necessary.

The volume problem: what satellites can and cannot tell you

Area change is measurable directly. Volume is not. Satellites observe the surface; the fill slope below the waterline is invisible to any passive optical sensor and to SAR. Volumetric estimates therefore require an assumption about the underwater geometry of the fill: typically a bathymetric survey of the pre-reclamation seabed combined with an assumed or modelled fill slope angle.

Where pre-construction bathymetry exists in the public record, or where a client can supply it, the volume calculation is straightforward: multiply the mapped area increment by the mean fill depth derived from the bathymetric model. Without that data, volume estimates carry substantial uncertainty, often 30–50% depending on how steeply the natural seabed shelved. Any analyst or vendor claiming precise volumetric figures without a bathymetric baseline is overstating what the physics allows.

Above-water surface models from stereo Pléiades Neo imagery can measure fill height above mean sea level to roughly 0.5 m vertical accuracy in favourable conditions. That constrains the above-water volume precisely but says nothing about the submerged portion, which is typically the larger fraction in shallow-water reclamation.

Legal and treaty-monitoring applications: what the archive proves

Satellite time-series have been cited in international arbitration and treaty-compliance assessments precisely because the archive is independent, continuous and difficult to dispute. A Landsat-to-Sentinel-2 chain provides dated evidence of land extent at roughly annual intervals from 1984 onwards, and at 5-day intervals from 2015. That temporal resolution is sufficient to attribute reclamation activity to specific periods, which matters when a state argues that construction predates a particular agreement or ruling.

The practical limit is spatial resolution. At 10–30 m, Sentinel-2 and Landsat cannot resolve whether a feature is a navigational aid, a temporary sand bar or the beginning of a permanent structure. That ambiguity is where very-high-resolution tasking, at 30–50 cm, becomes evidentially important. It is also where cloud cover creates gaps in the optical record that SAR must fill.

Satellize runs this kind of time-series analysis on open constellations and adds commercial tasking where boundary precision matters. The workflow is similar in structure to the crop-area estimation methodology used in the Tonga programme, adapted from spectral classification to shoreline change detection.

Decadal baselines and the 1984 problem

Landsat 5 TM began acquiring imagery in 1984. That archive, now freely available through the USGS Earth Explorer portal, is the only global satellite source that predates most of the large-scale reclamation projects now under scrutiny. Its 30 m resolution means that features smaller than roughly 60–90 m across are ambiguous, and the 16-day single-satellite revisit leaves seasonal gaps. But for establishing that a particular area of sea was open water in 1990 and is now land, it is authoritative.

Sentinel-2 data from 2015 onwards allows much finer temporal tracking. The combination of the two archives, joined by Landsat 7 and early Landsat 8 data in between, creates a continuous record. Processing that full chain consistently requires careful cross-sensor radiometric normalisation, since NDWI thresholds calibrated on Sentinel-2 do not transfer directly to Landsat 5 without adjustment. This is a solvable problem but one that requires explicit handling; analyses that ignore it will show apparent shoreline shifts that are artefacts of sensor differences rather than real change.

Delivery formats and honest detection limits

Reclamation change products are typically delivered as georeferenced polygon shapefiles or GeoTIFFs showing land-area increments per epoch, accompanied by area statistics and confidence intervals. For treaty or legal use, scene-level metadata, including acquisition time, tidal stage at acquisition and cloud-cover percentage, should accompany every mapped boundary.

The minimum detectable reclamation area at Sentinel-2 resolution is roughly one to four hectares, depending on the sharpness of the fill boundary and local water turbidity. At Planet SuperDove resolution, features as small as 0.1 hectares become detectable, though classification confidence drops in turbid or shallow water where the spectral contrast between fill and sea floor is reduced. Very-high-resolution stereo products from Pléiades Neo can resolve individual revetment blocks and produce surface models accurate to 0.5 m vertically above the waterline. Below the waterline, no passive optical system contributes anything without bathymetric support.

Typical figures

Coarse-resolution spatial baseline	30 m (Landsat 8/9); 10 m (Sentinel-2 NIR band)
Very-high-resolution boundary mapping	30 cm panchromatic, 1.2 m multispectral (Pléiades Neo)
Revisit frequency	5 days (Sentinel-2 dual satellite); daily (Planet SuperDove); 6 days (Sentinel-1 SAR)
Archive depth	1984 to present (Landsat); 2015 to present (Sentinel-2); 2014 to present (Sentinel-1)
Key spectral bands	Green and NIR for NDWI; SWIR for turbid-water discrimination; C-band SAR (5.4 GHz) for cloud-independent mapping
Minimum detectable reclamation area	1–4 ha at 10 m resolution; ~0.1 ha at 3–4 m resolution (Planet SuperDove)
Above-waterline vertical accuracy (stereo)	~0.5 m (Pléiades Neo stereo pair, favourable geometry)
Volumetric estimate uncertainty	30–50% without pre-construction bathymetry; reduces significantly with bathymetric baseline
Typical analytic latency	24–72 hours for archive-based change maps; 2–5 days for full time-series with tidal filtering
Delivery formats	GeoTIFF, shapefile/GeoJSON polygon layers, area-change CSV time-series, PDF evidence report with scene metadata

Analytics Satellize can run

Shoreline change time-series	NDWI thresholding applied to every cloud-free Sentinel-2 and Landsat scene; tidal-stage filtering using published tidal models	Dated polygon series (GeoJSON/shapefile) showing land-area increment per epoch, with area statistics and cloud-cover flags
Intertidal zone classification	Multi-temporal water-mask compositing across a full tidal cycle using Sentinel-1 SAR backscatter and Sentinel-2 optical data	GeoTIFF intertidal extent map distinguishing permanent land, intertidal sediment and open water
Reclamation activity phasing	SAR coherence change detection between successive Sentinel-1 pairs; low coherence flags active fill, rising coherence flags consolidation	Monthly activity-phase map (GeoTIFF) with active/consolidating/completed classification per polygon
Above-waterline surface model	Photogrammetric dense matching from Pléiades Neo stereo pairs; referenced to mean sea level via GCP or ICESat-2 lidar tie points	1 m resolution DSM (GeoTIFF) with vertical accuracy metadata; fill-height statistics by zone
Volumetric fill estimate	Area increment from optical time-series combined with client-supplied or publicly available pre-construction bathymetry; fill volume = area × mean depth from bathymetric model	Volume estimate table (CSV) with explicit uncertainty range; sensitivity analysis showing volume as a function of assumed fill slope
Treaty-period attribution report	Scene-by-scene Landsat and Sentinel-2 archive analysis with acquisition metadata; change events attributed to calendar periods aligned with legal milestones	PDF evidence report with annotated imagery, dated change polygons, tidal-stage and cloud-cover metadata per scene
Rapid-change alert	Automated NDWI difference detection on each new Sentinel-2 pass; threshold exceedance triggers alert	Email or API alert with GeoJSON polygon of newly detected land increment, typically within 24 hours of scene availability

Sources

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.