Urban infill plot detection for densification pipeline analysis
Very-high-resolution optical satellites can identify sub-threshold vacant gaps, rear-garden subdivisions and demolished footprints that never appear in official pipeline data, giving planners and investors an earlier, more granular read of densification pressure in land-constrained cities.
Sensors
- Maxar WorldView-3: 0.31 m panchromatic, 1.24 m multispectral (8 bands including coastal blue and SWIR). Single-pass revisit of 1 to 4.5 days depending on latitude and tasking priority. The sharpest commercially available optical sensor for resolving plot boundaries narrower than one metre.
- Airbus Pléiades Neo: 0.30 m panchromatic, 0.75 m multispectral (6 bands). Stereo and tri-stereo collection in a single pass, enabling surface-model differencing to confirm whether a cleared plot is truly flat or carries residual rubble. Revisit of 1 to 2 days across a four-satellite constellation.
- Planet SkySat: 0.50 m panchromatic, 0.72 m multispectral (5 bands). A 21-satellite fleet allows flexible tasking with collection windows of roughly 1 to 3 days in most urban areas. Slightly coarser than WorldView-3 or Pléiades Neo, so plots narrower than approximately 1.5 metres become ambiguous at this resolution.
- Sentinel-2 (ESA): 10 m multispectral (visible, NIR, SWIR). Free and open, with a 5-day revisit at the equator and better in mid-latitudes. Useless for individual infill plots but valuable as a temporal baseline for neighbourhood-level vegetation and impervious-surface change that contextualises the VHR findings.
Why the planning register is the wrong data source
Most national planning registers record applications above a minimum floor area or unit count. In England, for instance, permitted development rights allow certain demolitions, rear extensions and small residential conversions without any formal application at all. The result is a systematic gap between what is being built and what appears in the pipeline.
Infill development is particularly prone to this invisibility. A plot carved from a rear garden, a gap left by a demolished Victorian outbuilding, a cleared side return between two semis: none of these reliably triggers a data event that aggregators can capture. Satellite imagery does not care about reporting thresholds. A cleared footprint is a cleared footprint, and at 0.3 m resolution it is detectable.
What a cleared footprint actually looks like from 500 km
At WorldView-3 or Pléiades Neo resolution, a demolished structure leaves a distinctive spectral and textural signature. Bare aggregate, disturbed soil and concrete rubble have higher reflectance in the red and SWIR bands than the surrounding rooftop material. Texture metrics derived from the panchromatic band, particularly grey-level co-occurrence matrix (GLCM) measures of homogeneity and contrast, distinguish a flat cleared plot from a standing roof even when the two occupy similar spectral space.
Rear gardens that have been subdivided and cleared show a different pattern: a sharp rectilinear boundary appears within a previously continuous vegetation polygon, often accompanied by the removal of tree canopy. NDVI differencing between two acquisition dates flags the vegetation loss; the geometry of the new boundary, extracted by building-footprint differencing against a cadastral or prior-image baseline, confirms subdivision rather than simple garden clearance.
Honest caveat: dense deciduous tree canopy is a genuine problem. A plot obscured by a mature tree crown at leaf-on acquisition will not be detectable without a leaf-off image or a SAR complement. Plots narrower than roughly two metres also fall below reliable detection at 0.5 m resolution and become ambiguous even at 0.3 m when shadow from adjacent buildings compounds the geometry.
The differencing workflow and its inputs
The core method is building-footprint differencing: a reference footprint layer (from a prior VHR acquisition, OpenStreetMap, Ordnance Survey or a national cadastre) is compared against a freshly derived footprint extracted from a new image using object-based image analysis (OBIA) or a trained convolutional segmentation model. Polygons present in the reference and absent in the new image flag demolitions. Polygons in the new image absent from the reference flag construction starts, which is a different use case covered elsewhere in this library.
The quality of the reference layer matters enormously. A cadastral dataset that is three years out of date will generate false positives at sites that were already cleared before the analysis window. Where a high-quality reference is unavailable, a two-date VHR stack from the archive serves as its own baseline, though archive imagery older than roughly five years may predate the densification pressure being studied. Archive depth for WorldView-3 runs to 2014; Pléiades Neo to 2021.
Cloud cover is the operational constraint that does not disappear with money. Persistent cloud over UK cities in winter can delay a cloud-free acquisition by several weeks even with aggressive multi-satellite tasking. Scheduling collections across leaf-off and cloud-minimised windows, typically October to March in northern Europe, is standard practice.
Turning detections into a pipeline signal
A single detected cleared plot is an observation. A density map of cleared plots across a local authority area, updated quarterly, is a market signal. The analytical value comes from aggregation: how many sub-threshold plots have been cleared in a given ward over the past 12 months, and how does that compare to the prior period and to adjacent wards?
Overlaying detections against planning-application data reveals the gap between what the register shows and what the imagery shows. That gap is the undercounted pipeline. For a residential developer or a housing association assessing land supply in a constrained borough, that number is more useful than the headline consented-units figure.
Detections can also be filtered by plot area, proximity to existing transport nodes, or flood-zone exclusion to produce a ranked opportunity list. This is not a substitute for site-level due diligence, but it narrows the search space before ground-truthing is required. Satellize runs this kind of differencing workflow on client-specified areas of interest, outputting GIS layers and summary statistics rather than raw imagery.
Resolution floors, revisit gaps and other honest limits
The 0.3 m sensors are the right tool for this use case, but they are not magic. Positional accuracy of VHR imagery after orthorectification is typically 3 to 5 m CE90 without ground control points, and 0.5 to 1 m CE90 with them. For plot-boundary work in dense urban areas, ground control or a rigorous tie-point registration to a known cadastral layer is not optional.
Multispectral bands on WorldView-3 and Pléiades Neo are collected at 1.24 m and 0.75 m respectively, not at the panchromatic resolution. Pan-sharpening recovers spatial detail but does not recover spectral fidelity at sub-metre scale. Analyses that depend on precise NDVI thresholds should account for the pan-sharpening artefacts that appear at high-contrast edges, particularly roofline-to-garden transitions.
Finally, the method detects physical change, not legal change. A cleared plot may have planning permission already granted, may be under an option agreement, or may be a simple garden clearance with no development intent. Ground-truthing a sample of detections against the planning register and Land Registry is essential before acting on any individual site.
Typical figures
| Best available spatial resolution (pan) | 0.30 m (Pléiades Neo, WorldView-3) |
| Multispectral resolution | 0.75 m (Pléiades Neo) to 1.24 m (WorldView-3) native; pan-sharpened to pan resolution |
| Revisit (tasked VHR) | 1 to 4.5 days depending on sensor and latitude; cloud-free acquisition may take several weeks in persistently overcast climates |
| Minimum reliably detectable plot width | Approximately 2 m at 0.3 m resolution with favourable shadow geometry; ambiguous below that threshold |
| Positional accuracy (orthorectified, with GCPs) | 0.5 to 1 m CE90; 3 to 5 m CE90 without ground control |
| Archive depth | WorldView-3 from 2014; Pléiades Neo from 2021; Pléiades-1 (predecessor, 0.5 m) from 2011 |
| Spectral bands relevant to this use case | Panchromatic, Red, NIR (NDVI), SWIR (WorldView-3 only, useful for bare-soil discrimination) |
| Typical area per tasked collect | 25 km² per strip (WorldView-3); up to 400 km² per day with SkySat fleet |
| Delivery formats | GeoTIFF (orthorectified), GeoJSON / Shapefile (footprint polygons), CSV (plot-level attributes), QGIS / ArcGIS compatible |
| Cloud cover limit for usable imagery | Typically less than 10 % cloud cover over the area of interest required for reliable detection |
Analytics Satellize can run
| Cleared-footprint detection layer | Building-footprint differencing using OBIA segmentation on two-date VHR stack; GLCM texture metrics on panchromatic band to distinguish bare ground from standing structure | GeoJSON polygon layer of detected cleared plots with area, centroid coordinates, confidence score and detection date |
| Rear-garden subdivision map | NDVI differencing (Sentinel-2 or VHR multispectral) to flag vegetation loss; rectilinear boundary geometry check against prior garden polygon to confirm subdivision rather than incidental clearance | GeoJSON layer of probable subdivision events, attributed with plot area and change date range |
| Undercounted pipeline estimate | Spatial join of satellite-detected clearances against local authority planning register; gap analysis quantifying detections with no corresponding application | Ward-level summary table (CSV and PDF report) showing detected clearances, registered applications and estimated unregistered pipeline |
| Ranked opportunity list | Multi-criteria scoring of detected plots by area, distance to public transport nodes, flood-zone exclusion and prior land-use class | Ranked CSV or GIS layer of candidate infill sites for developer or housing-association review |
| Quarterly change bulletin | Repeat tasking on a defined area of interest; differencing against prior quarter's baseline; trend statistics on clearance rate by ward | Quarterly PDF bulletin with map outputs and time-series chart of clearance activity |
| Stereo-derived surface model for rubble confirmation | Pléiades Neo tri-stereo collection processed to a 0.5 m DSM; differencing against prior DSM or LiDAR reference to confirm ground-level clearance versus partial demolition | GeoTIFF DSM difference raster and attributed polygon layer flagging plots with residual above-ground material |
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.