Land parcel change detection for cadastral update
Cadastral records drift from reality through subdivision, encroachment and informal construction. Sub-30 cm satellite imagery, overlaid against legal vector boundaries, can flag discrepancies before they become disputes.
Sensors
- Maxar WorldView-3: 31 cm panchromatic GSD at nadir, 1.24 m multispectral (8 bands including SWIR), revisit roughly 1 day at mid-latitudes. The panchromatic channel resolves fence posts, kerb lines and narrow access strips that define cadastral boundaries in practice.
- Airbus Pléiades Neo: 30 cm panchromatic GSD, 1.2 m multispectral (6 bands), constellation of two satellites giving sub-daily revisit over priority sites. Tri-stereo collection mode supports 50 cm digital surface models useful for detecting raised structures inside parcel boundaries.
- Planet SkySat: 50 cm panchromatic GSD, 1 m multispectral. Revisit can be tasked to daily frequency. Slightly coarser than WorldView-3 or Pléiades Neo but adequate for flagging large encroachments and amalgamations; less reliable for resolving sub-metre boundary disputes.
- Airbus SPOT 7: 1.5 m panchromatic GSD, 6 m multispectral. Useful for screening large cadastral areas to prioritise which parcels warrant VHR follow-up, but insufficient on its own for boundary-line resolution. Archive depth extends to 2014.
Why cadastral records age badly
A cadastral database is a legal document that describes the world as it was at the moment of survey. The world does not stay still. Owners subdivide informally, neighbours erect walls a metre into the adjacent parcel, access roads are absorbed into private plots, and agricultural land is quietly amalgamated without formal registration. In many jurisdictions the gap between the legal record and the physical reality widens for years before anyone notices, and when they do notice it is usually because a sale, mortgage or compulsory acquisition has forced a survey.
Satellite change detection does not replace the licensed surveyor who must certify a boundary. What it does is identify, at scale and low cost, which parcels in a registry area have changed enough to warrant a ground survey. That triage function is where the economics are compelling: a national land authority cannot resurvey every parcel every year, but it can afford to screen every parcel from orbit and dispatch surveyors only where the imagery flags a discrepancy.
The 30 cm threshold and why it matters
Positional accuracy in VHR satellite imagery is governed by two separate quantities: ground sample distance (GSD) and absolute geolocation accuracy. GSD determines the smallest feature that can be resolved; geolocation accuracy determines how well the image aligns to a coordinate reference system. WorldView-3 and Pléiades Neo both achieve 30 cm GSD and, with rational polynomial coefficient (RPC) refinement against ground control points, can reach sub-50 cm circular error probable (CEP) in absolute position. That combination is what makes sub-metre boundary work tractable.
The practical rule is that a boundary discrepancy needs to be at least two to three pixels wide before it can be detected reliably. At 30 cm GSD that means encroachments of roughly 60 to 90 cm become detectable. Smaller offsets fall within the noise of image registration and should not be reported as definitive findings. This is an honest limit of the method, and any workflow that claims to resolve 10 cm cadastral disputes from current commercial satellites is overstating the evidence. For disputes at that precision, differential GNSS or terrestrial LiDAR remains the only reliable instrument.
Object-based analysis and edge detection: what the algorithms actually do
Pixel-based classification struggles with cadastral work because a boundary is a line, not an area. The productive approaches are object-based image analysis (OBIA) and edge detection, often combined. In OBIA the image is first segmented into spectrally and geometrically coherent objects, fences, walls, road edges, vegetation margins, then each object is compared against the registered boundary vector. Discrepancies beyond a configurable tolerance trigger a flag. The tolerance must account for image registration uncertainty; setting it too tight generates false positives from registration noise, too loose and real encroachments are missed.
Edge detection methods, including Canny and Sobel operators applied to the panchromatic band, extract linear features from the image independently of the cadastral layer. Those extracted lines are then compared with registered boundary geometry using a Hausdorff distance metric or similar. Where the detected edge diverges from the registered line by more than the positional uncertainty budget, the system marks a candidate change. Both approaches produce a ranked list of candidate discrepancies rather than a binary pass or fail, which is the appropriate output given the inherent uncertainty.
Temporal change detection adds a second layer. Differencing two orthorectified images from different epochs, one pre-registered as baseline, one current, highlights areas of radiometric change inside or across parcel boundaries. New impervious surfaces, cleared vegetation and fresh construction all produce strong difference signals. The method is sensitive to seasonal variation in vegetation, so phenologically matched image pairs reduce false positives significantly.
Orthorectification, registration and the error budget
The entire workflow depends on the imagery and the cadastral vector dataset living in the same coordinate reference system to a known accuracy. Cadastral datasets in many countries were digitised from paper maps at scales of 1:2500 or coarser, introducing inherent positional uncertainty that can exceed one metre even before the satellite image enters the picture. Analysts must characterise the cadastral data's own accuracy before interpreting image-to-boundary discrepancies. A one-metre offset between an observed wall and a registered boundary line means nothing if the cadastral dataset itself has one-metre uncertainty.
Rigorous orthorectification using a digital elevation model (DEM) and RPC refinement is non-negotiable for boundary work. In flat terrain the DEM contribution is minor, but in hilly or mountainous areas a one-degree slope can shift a 30 cm GSD image feature by several metres if a coarse DEM is used. The Copernicus DEM at one arc-second (roughly 30 m posting) is the standard open reference; higher-density DEMs derived from stereo VHR collection improve results further in complex terrain.
Subdivision, amalgamation and encroachment: three distinct signatures
Subdivision leaves a characteristic signature: a new internal linear feature, typically a wall, fence or access track, appears inside a previously undivided parcel. The parent parcel boundary remains unchanged but the interior now contains a new edge. Amalgamation is harder to detect because it involves the removal of a boundary feature. A fence that disappears between epochs is a weaker signal than one that appears, and vegetation regrowth can obscure the former line. Encroachment sits between the two: a structure or land-use change crosses the registered boundary into an adjacent parcel, often incrementally over multiple image epochs.
Each type requires a slightly different detection logic. Subdivision detection is well suited to within-parcel edge extraction. Amalgamation detection benefits from change-pair differencing to identify where previously detectable linear features have vanished. Encroachment detection is most reliably done by buffering the registered boundary line and examining the buffer zone for new impervious or structural features in the current image that were absent in the baseline. Satellize applies these detection classes as separate analytic layers, which allows a land authority to filter alerts by change type rather than receiving an undifferentiated list of flagged parcels.
Delivery, integration and what satellite data cannot do
The output of a cadastral change-detection run is a prioritised alert layer: a GIS polygon or line dataset in which each flagged feature carries a confidence score, a change-type classification and the image date pair from which it was derived. That layer overlays directly onto a land authority's existing cadastral GIS without requiring a new platform. Alerts above a configurable confidence threshold can trigger automatic workflow tickets in cadastral management systems via a simple API feed.
The limits are real and should be stated plainly. Satellite imagery cannot determine legal intent. A wall that appears to cross a boundary might reflect a surveying error in the original cadastral dataset, an agreed boundary adjustment that was never formally registered, or a genuine encroachment. The satellite flags the physical discrepancy; a human, and eventually a licensed surveyor, must determine the legal consequence. Imagery also cannot see through dense tree canopy, which is a genuine problem in forested rural cadastres. For those areas, airborne LiDAR or synthetic aperture radar coherence methods are better tools, and a combined approach is more informative than either alone.
Typical figures
| Finest available GSD (panchromatic) | 30 cm (WorldView-3, Pléiades Neo) |
| Absolute geolocation accuracy (with GCP refinement) | Sub-50 cm CEP typical for WorldView-3 and Pléiades Neo |
| Minimum detectable boundary discrepancy | Approximately 60–90 cm at 30 cm GSD (2–3 pixel rule); sub-metre disputes require ground survey |
| Revisit frequency (tasked VHR) | Sub-daily to 1–2 days (WorldView-3, Pléiades Neo); daily on demand (SkySat) |
| Multispectral bands | 4–8 bands visible to SWIR (WorldView-3); 4–6 bands visible to NIR (Pléiades Neo, SkySat) |
| DEM reference for orthorectification | Copernicus DEM GLO-30 (1 arc-second posting) as standard open baseline |
| Archive depth | WorldView-3 from 2014; Pléiades from 2012; SPOT 7 from 2014 |
| Delivery format | GeoPackage, GeoJSON or Shapefile alert layer; COG GeoTIFF imagery; optional API feed |
| Latency from tasking to delivered alert layer | Typically 24–72 hours after image acquisition, depending on processing queue and area size |
Analytics Satellize can run
| Parcel-level change alert layer | Temporal image differencing with thresholded change mask, registered to cadastral vector | GIS polygon layer with confidence score, change epoch and change-type flag per parcel |
| Boundary edge discrepancy map | Canny or Sobel edge detection on panchromatic band; Hausdorff distance comparison against registered boundary lines | Line GIS layer showing detected edges and their offset distance from registered cadastral boundary |
| Subdivision detection report | OBIA segmentation to detect new internal linear features within registered parcel polygons | Ranked list of parcels with probable new internal boundaries, with supporting image chip |
| Encroachment buffer analysis | Boundary-buffer zone inspection for new impervious surface or structural features across registered parcel limits | Alert polygons clipped to buffer zone with estimated encroachment area in square metres |
| Baseline orthorectified image archive | RPC-based orthorectification refined against Copernicus DEM GLO-30 and available GCPs | Cloud-optimised GeoTIFF image stack per area of interest, co-registered across epochs |
| Cadastral data quality assessment | Statistical comparison of cadastral vector positional accuracy against image-derived control features | Report quantifying cadastral dataset uncertainty by zone, informing detection confidence thresholds |
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.