Unauthorised construction and land encroachment detection
Time-series optical and SAR imagery can flag unauthorised structures, rights-of-way encroachments, and exclusion-zone violations weeks before a ground inspector arrives. Resolution, cloud cover, and seasonal vegetation all shape what the data can and cannot prove.
Sensors
- Planet SuperDove: 3 m ground sample distance, 8-band multispectral (coastal blue through NIR-II), near-daily global revisit from a constellation of roughly 200 satellites. The combination of daily cadence and sub-5 m resolution makes it the primary sensor for detecting small structures quickly after they are built.
- Sentinel-1 SAR (C-band, ESA): Interferometric Wide swath mode at 10 m resolution, 6-day repeat at mid-latitudes (12-day at the equator). Cloud-transparent and night-capable. Coherence change detection on Sentinel-1 pairs can flag new impervious surfaces even when optical imagery is obscured for weeks.
- Maxar WorldView Legion: 30 cm native resolution, up to 15 revisits per day over priority areas. Used for confirmation and legal-grade documentation once a candidate encroachment has been flagged by coarser sensors. Not a monitoring tool at scale; a verification tool for specific parcels.
- Sentinel-2 MSI (ESA): 10 m in visible and NIR bands, 5-day revisit with two satellites. Free and open. Useful for monitoring large exclusion zones such as airport approach surfaces or wide pipeline corridors, though its 10 m floor means structures smaller than roughly 30 m² are unreliable detections.
What a new roof looks like to a satellite
Change detection is conceptually simple: compare two images of the same place taken at different times and find pixels that have shifted in reflectance or radar backscatter. In practice, the difficulty lies in separating genuine construction from the many other things that change a pixel, including crop growth, soil moisture, shadow angle, and atmospheric haze.
Planet SuperDove's 3 m resolution means a structure of roughly 9 m² occupies a single pixel. Reliable detection in practice requires a target to span several contiguous pixels with a consistent spectral signature, so the operational minimum detectable structure is closer to 50 to 100 m², depending on roof material and surrounding land cover. Concrete and metal roofing have high NIR reflectance and low NDVI, which separates them cleanly from vegetation. Thatch or earth-coloured materials are harder. Sentinel-1 coherence loss is less sensitive to material but picks up any surface that disrupts the phase relationship between two passes, making it useful for detecting earthworks and foundations before a roof is even in place.
Seasonal vegetation is the principal source of false positives
In tropical and subtropical regions, the transition from dry season to wet season can double the NDVI of a grass-covered plot within a fortnight. A naive pixel-differencing algorithm will flag those plots as changed, because they are changed, just not in the way enforcement agencies care about. The standard mitigation is to compare images acquired in the same phenological season across different years rather than consecutive months, and to apply a vegetation mask derived from NDVI thresholds before running the structural change classifier.
Even with seasonal normalisation, false-positive rates in peri-urban areas with mixed agriculture and informal settlement typically run between 5 % and 20 % at the pixel level before post-processing filters are applied. Reducing that to operationally acceptable levels (under 5 % false positives at the parcel level) generally requires a second-pass classifier trained on local land-cover classes, plus a minimum persistence criterion: a change must appear in at least two independent acquisitions before it is elevated to an alert. This adds latency of roughly one to two weeks but substantially improves precision.
Exclusion zones are not all the same problem
Airport approach surfaces, pipeline easements, transmission-line corridors, and coastal setback zones each have different geometry and different legal definitions of what constitutes a violation. Approach surfaces are three-dimensional: a structure that is legal at ground level may penetrate the obstacle limitation surface at its rooftop. Satellite imagery measures planimetric footprint reliably but height only indirectly, through shadow length and stereo parallax in high-resolution imagery. WorldView Legion stereo can produce digital surface models with vertical accuracy of roughly 1 to 2 m under good conditions, which is sufficient to assess whether a structure breaches a 1:50 or 1:40 glide-slope surface. Pipeline easements are a planimetric problem: any structure whose footprint intersects the buffer polygon is a candidate violation, regardless of height.
Transmission-line corridors add a vegetation-encroachment dimension that is the inverse of the false-positive problem described above: here, vegetation growth is the encroachment. Tree canopy advancing into a right-of-way is detectable in Sentinel-2 and Planet imagery through canopy height proxies and shadow analysis, though precise height measurement again requires lidar or stereo.
Legal admissibility: what the evidence chain requires
Satellite imagery has been admitted as documentary evidence in land-dispute and planning-enforcement proceedings in a number of jurisdictions, including the UK, Australia, Kenya, and several US states, but the evidentiary requirements vary considerably. Courts and tribunals generally ask three questions: can the image be authenticated as unaltered? Can the acquisition time and position be independently verified? And is the resolution sufficient to support the specific claim being made?
Commercial providers including Planet and Maxar can supply chain-of-custody metadata, sensor calibration records, and in some cases expert witness support. Sentinel data carries ESA provenance documentation. The weaker link is usually the analysis layer: a GIS polygon drawn by an algorithm carries less inherent credibility than one drawn by a licensed surveyor using the satellite image as a backdrop. In jurisdictions where enforcement is anticipated, the workflow should separate the detection step (algorithmic, for speed and scale) from the documentation step (human-verified, for legal weight). Presenting raw pixel coordinates to a planning tribunal without a qualified analyst's interpretation is a common and avoidable mistake.
Revisit cadence determines how much construction you miss
A structure can go from bare earth to roofed shell in under two weeks in many construction traditions. With a 6-day Sentinel-1 revisit, the maximum undetected construction window is six days at mid-latitudes, though cloud cover over the optical archive during that window means the SAR evidence may be the only contemporaneous record. Planet's near-daily revisit shrinks the window further but is subject to cloud and off-nadir angle variation that reduces effective clear-sky coverage to perhaps 70 to 80 % of days over persistently cloudy regions.
For enforcement purposes, the date of first detection is not the same as the date of construction. Analysts should report the date of the earliest image showing the structure and the date of the latest image showing its absence, giving an honest construction window rather than a spuriously precise date. This distinction matters in legal proceedings where the timing of a permit application or an enforcement notice is disputed.
From alert to enforcement file
The practical workflow runs in three stages. First, a continuous change-detection monitor over the defined zone of interest flags candidate parcels, typically delivered as a GIS alert layer updated on each new acquisition. Second, a human analyst reviews flagged parcels against cadastral boundaries, permit databases, and historical imagery to filter false positives and classify the violation type. Third, for confirmed cases, a high-resolution tasking order on WorldView Legion or a comparable commercial sensor produces the documentation-grade imagery needed for an enforcement file.
Satellize structures analytics pipelines across open and commercial constellations for clients who need this three-stage workflow without building it in-house. The approach is the same one applied in the Tonga crop-estimation programme: open-constellation monitoring at scale, commercial tasking reserved for the specific sites and moments that justify the cost. Clients receive a parcel-level alert feed, a monthly summary report, and, on request, imagery packages formatted to meet the evidentiary standards of their jurisdiction's planning tribunal.
Typical figures
| Primary optical resolution | 3 m (Planet SuperDove); 10 m (Sentinel-2); 30 cm (WorldView Legion, verification only) |
| Primary SAR resolution | 10 m (Sentinel-1 IW mode, C-band) |
| Revisit cadence | Near-daily (Planet); 5–6 days (Sentinel-2 / Sentinel-1 at mid-latitudes); up to 15×/day (WorldView Legion over priority areas) |
| Minimum detectable structure (operational) | ~50–100 m² for optical change detection at 3 m resolution; larger structures more reliable; SAR coherence can detect earthworks below this threshold |
| Typical false-positive rate (parcel level) | 5–20 % before post-processing; reducible to <5 % with seasonal normalisation and persistence filters |
| Alert latency | 1–3 days from acquisition for automated flag; 1–2 additional weeks if persistence criterion applied |
| Archive depth | Sentinel-1 and Sentinel-2 from 2014–2015; Planet from 2016; Maxar archive from early 2000s (variable coverage) |
| Cloud limitation | Optical sensors affected; SAR (Sentinel-1) cloud-transparent. Effective clear-sky optical coverage ~70–80 % of days in persistently cloudy regions |
| Vertical height estimation | 1–2 m vertical accuracy from WorldView stereo DSM under good conditions; shadow-based height indirect and lower accuracy |
| Delivery formats | GeoTIFF change masks, GeoJSON/Shapefile parcel alert layers, PDF enforcement report, imagery packages with provenance metadata |
Analytics Satellize can run
| Continuous zone-of-interest change monitor | Bi-temporal and multi-temporal pixel differencing with NDVI-based vegetation masking and seasonal normalisation; applied to Planet SuperDove and Sentinel-2 time series | GeoJSON alert layer updated per acquisition cycle, flagging candidate new-structure pixels with confidence score and acquisition dates |
| SAR coherence change detection | Sentinel-1 coherence differencing between sequential 6-day pairs; coherence loss identifies new impervious surfaces and earthworks independent of cloud cover | Raster coherence-change layer clipped to zone of interest, delivered as GeoTIFF with acquisition metadata |
| Parcel-level encroachment classification | Intersection of detected change polygons with cadastral boundary and rights-of-way GIS layers; violation type classified by geometry (footprint overlap, approach-surface penetration, corridor intrusion) | Attributed Shapefile or GeoPackage with violation type, first-detection date, construction window, and confidence rating per parcel |
| Verification imagery package | Commercial tasking of WorldView Legion at 30 cm resolution over confirmed candidate parcels; stereo pair acquisition for DSM generation where height assessment is required | Orthorectified GeoTIFF with chain-of-custody metadata, optional DSM, formatted for planning-tribunal submission |
| Historical baseline and construction-date bounding | Archive search across Sentinel, Planet, and Maxar catalogues to establish the latest image showing absence and earliest image showing presence of a structure; reported as a construction window, not a point date | Chronological image strip with annotated before/after pair and written construction-window statement suitable for legal documentation |
| Monthly enforcement summary report | Aggregation of parcel-level alerts over the reporting period; trend analysis of encroachment rate and spatial distribution within the zone of interest | PDF report with maps, parcel inventory table, and analyst commentary; suitable for regulatory authority or project-finance covenant reporting |
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.