Sensors
- Pleiades Neo (Airbus): 30 cm native resolution in panchromatic, 1.2 m multispectral (four bands plus red-edge and deep blue on Neo). Constellation of two satellites achieves same-day revisit over most cities. The resolution is sufficient to resolve individual structures as small as 2–3 m across, making rear-garden outbuildings clearly distinguishable from vegetation.
- WorldView Legion (Maxar): 30 cm panchromatic, up to 15 revisits per day over targeted cities at mid-latitudes. High revisit frequency is the key contribution: it allows change detection between epochs separated by days rather than months, catching construction in progress rather than after completion.
- SkySat (Planet Labs): 50 cm panchromatic, 72 cm multispectral. The constellation of 21 satellites can revisit a city multiple times daily. Resolution is slightly coarser than Pleiades Neo or WorldView Legion but adequate for detecting new impervious surfaces and structure footprints above roughly 10–15 m².
- WorldView-3 (Maxar): 31 cm panchromatic, 1.24 m eight-band multispectral, plus 16-band SWIR at 3.7 m. The SWIR bands help distinguish fresh concrete and new roofing materials from soil or vegetation in spectral classification, adding a material-change signal to the geometric one.
Why the land register and the ground diverge
Cadastral records are legal instruments, not physical surveys. They record the boundary agreed at the time of registration and are updated only when a transaction triggers a formal resurvey. In practice, owners subdivide plots, build on rear gardens, and erect secondary dwellings for years before any registration event forces an update. In fast-growing cities across the Global South and in peri-urban Europe, the gap between the registered parcel geometry and the actual pattern of occupation can span a decade or more.
The consequence for local government is compounding: property tax rolls undercount the actual building stock, infrastructure planning is based on stale density figures, and building-code enforcement has no visibility of structures that were never permitted. The problem is not dramatic illegal settlement on greenfield land (that is a different detection challenge) but something quieter and more pervasive: the incremental filling of space within already-registered parcels.
What a floating roof gives away
A new structure inside a rear garden appears in sub-metre imagery as three simultaneous signals: a new impervious surface where there was previously bare soil or vegetation, a shadow cast by a new vertical element, and a spectral shift in the pixel cluster. None of these signals requires the analyst to know where the cadastral boundary runs. The change is detected geometrically by differencing building-footprint rasters between two epochs, then cross-referenced against the registered parcel layer to flag cases where new footprints fall inside a single registered plot.
The detection floor is set by sensor resolution and the minimum mappable unit. At 30 cm (Pleiades Neo or WorldView Legion), structures with a footprint above roughly 9–12 m² are reliably detected in good illumination conditions. SkySat at 50 cm raises that floor to approximately 20–25 m². Cloud is the primary operational constraint: optical sensors cannot see through it, and in tropical cities with persistent cloud cover, the revisit advantage of WorldView Legion or SkySat is the practical answer, accumulating cloud-free acquisitions faster than a lower-cadence sensor would.
Shadow geometry also carries information about building height. With a known solar elevation angle at acquisition time, the length of a cast shadow gives a first-order estimate of wall height. This is not a substitute for stereo-derived height models, but it is a useful screening signal when stereo pairs are not available.
Building the change signal: epochs, differencing, and false-alarm control
The analytical workflow starts with co-registered image pairs from two epochs. Geometric co-registration to sub-pixel accuracy is non-negotiable: at 30 cm resolution, a one-pixel misalignment introduces a 30 cm positional error that generates spurious change detections along every building edge. Once co-registered, building footprints are extracted from each epoch using a combination of morphological filtering and, increasingly, convolutional neural network segmentation trained on labelled urban imagery. The difference between epoch-one and epoch-two footprint rasters produces a candidate change layer.
False alarms come from seasonal vegetation change (a tree in leaf obscuring or revealing a structure), temporary objects (parked vehicles, construction equipment), and radiometric differences between acquisitions taken under different sun angles or atmospheric conditions. Filtering uses a minimum area threshold, a persistence check across at least two independent acquisitions in the later epoch, and a spectral plausibility test confirming that the new pixel cluster has the reflectance signature of a built material rather than vegetation or bare soil. After filtering, the remaining detections are overlaid on the cadastral parcel layer. Those that fall within a single registered parcel, away from the parcel boundary, are flagged as probable infill densification.
Honest limits of the method
Sub-metre commercial imagery is not free. Tasking costs and data-licensing fees mean that systematic city-wide monitoring at monthly or quarterly cadence is expensive, and the economics typically push towards risk-based targeting: survey the parcels most likely to have changed based on permit applications, neighbourhood growth rates, or prior detections, rather than blanketing the entire city.
The method detects physical change, not legal status. A new structure inside a registered parcel might be a permitted extension, a temporary greenhouse, or a utility shed, as well as an unregistered dwelling. Ground-truth confirmation or cross-referencing with the building-permit database is required before any enforcement action. The satellite flags the anomaly; a human or a permit-record query resolves the ambiguity.
Flat rooftops on multi-storey buildings present a specific problem. A new structure built on top of an existing building may be invisible from nadir-looking imagery if the parapet is high enough. Oblique acquisition modes, available on Pleiades Neo and WorldView-3, partially address this, but rooftop additions remain harder to detect than ground-level infill.
Fitting this into a land-administration workflow
The most practical integration point is the field-inspection queue. Rather than sending inspectors to every parcel on a rotating schedule, the satellite-derived change layer becomes the prioritisation input: inspectors visit the flagged parcels first. This concentrates enforcement effort where the probability of finding an unregistered structure is highest, and it creates an audit trail showing that the inspection was triggered by an objective, repeatable signal rather than a complaint or a discretionary decision.
For municipalities running a property-tax revaluation, the same change layer doubles as a stock-update tool. New footprints identified since the last valuation cycle are added to the assessment roll, with floor-area estimated from the footprint and a locally calibrated height proxy. Satellize has applied analogous multi-date footprint-differencing logic in its analytics work; the Tonga crop-estimation programme used comparable change-detection principles on agricultural parcels, and the same epoch-differencing architecture transfers directly to urban built-environment monitoring.
Archive depth matters for retrospective analysis. Pleiades (the earlier generation) has tasked imagery over many cities since 2012, and WorldView-2 and -3 archives extend back to 2009 and 2014 respectively. A municipality that wants to reconstruct a decade of infill activity to support a regularisation programme can often do so from commercial archive, without commissioning new tasking.
Typical figures
| Best available spatial resolution | 30 cm panchromatic (Pleiades Neo, WorldView Legion, WorldView-3) |
| Multispectral resolution | 1.2–1.24 m (Pleiades Neo, WorldView-3 MS); 72 cm (SkySat) |
| Revisit frequency | Up to 15 times/day over targeted cities (WorldView Legion); 1–2 times/day typical for Pleiades Neo; multiple times/day for SkySat over tasked areas |
| Minimum detectable new footprint | Approximately 9–12 m² at 30 cm; approximately 20–25 m² at 50 cm, under good illumination and cloud-free conditions |
| Cloud penetration | None. All sensors listed are optical. High revisit cadence is the operational mitigation. |
| Spectral bands | Panchromatic plus 4-band multispectral (standard); red-edge and deep blue (Pleiades Neo); 8-band VNIR + 16-band SWIR (WorldView-3) |
| Archive depth | WorldView-2 from 2009; WorldView-3 from 2014; Pleiades from 2012; SkySat from approximately 2015 |
| Positional accuracy (after orthorectification) | Typically 0.3–1.0 m CE90 with ground control; sub-pixel co-registration required for change detection |
| Delivery formats | GeoTIFF orthoimage, GeoPackage or Shapefile change-polygon layer, CSV parcel-flag report |
Analytics Satellize can run
| New-footprint change layer | Multi-date building-footprint extraction via CNN segmentation followed by binary raster differencing and morphological filtering | GeoPackage polygon layer of new structures detected between two specified epochs, attributed with estimated footprint area and detection confidence score |
| Parcel-level densification flag | Spatial join of new-footprint layer against registered cadastral parcel boundaries; flags parcels where new footprints appear interior to a single registered plot | CSV or GIS layer of flagged parcel IDs with coordinates, flagged area, and epoch pair, formatted for import into land-administration software |
| Impervious-cover change ratio per parcel | Spectral index classification (normalised difference built-up index or similar) applied to multispectral imagery at both epochs; impervious fraction computed per parcel polygon | Tabular report of impervious-cover percentage at epoch one and epoch two per parcel, with delta value, suitable for property-tax revaluation input |
| Building-height proxy from shadow length | Shadow extraction from panchromatic imagery combined with solar elevation angle at acquisition time to estimate wall height of new structures | Attribute field appended to new-footprint layer giving estimated height range (typically ±0.5 m accuracy) for each detected structure |
| Inspector prioritisation queue | Ranked list of flagged parcels ordered by detection confidence, estimated new floor area, and time since last known inspection | Spreadsheet or API feed compatible with field-inspection management systems, updated at each new imagery epoch |
| Retrospective densification timeline | Archive time-series analysis across available commercial imagery epochs to reconstruct year-by-year infill history for a defined study area | Animated GIS layer and summary chart showing cumulative new footprint area per year, suitable for regularisation programme evidence packs |
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.