Sensors
- Planet SkySat: Delivers 50 cm pan-sharpened optical imagery with tasking latency of roughly 24 hours and revisit on demand for a given area of interest. At that resolution, foundation slabs, scaffolding and material stockpiles are individually legible. Cloud cover remains a hard constraint; persistent overcast can block a site for weeks.
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral, with eight SWIR bands that can distinguish bare concrete, disturbed soil and standing water on a site. Revisit at mid-latitudes is roughly 1 to 4.5 days depending on off-nadir tasking. Archive depth extends to 2014, enabling baseline establishment for change detection.
- Sentinel-1 SAR (C-band, coherence): Freely available at 10 m resolution in Interferometric Wide Swath mode, with a 6-day repeat at the equator and 12-day where a single satellite covers. Coherence loss between two passes reliably flags ground disturbance from excavation, demolition or material movement even through cloud, smoke and at night. It cannot resolve individual structures but is an excellent screening layer across large urban extents.
- Google Open Buildings: A published dataset of building footprints derived from satellite imagery, covering large parts of Africa, South and Southeast Asia and Latin America. Useful as a baseline against which new footprints detected in later imagery can be differenced to estimate unit counts and gross floor area growth. Coverage and update cadence vary by region; it is not a substitute for fresh tasking in fast-changing areas.
Why planning-permission data fails operators in secondary cities
In established markets, network planners can cross-reference local-authority building registers against subscriber forecasts. In the secondary cities of sub-Saharan Africa, South and Southeast Asia, and parts of Latin America, that register either does not exist, is months out of date, or is not publicly accessible. Informal and semi-formal construction, which accounts for a substantial share of new housing stock in these regions, never appears in it at all.
The consequence is that operators discover new demand only after activation. A tower serving a district of 8,000 people is suddenly asked to serve 12,000. Capacity upgrades that take three to nine months to procure and install are ordered reactively. Satellite-based construction monitoring inverts that sequence.
What a disturbed-earth signature gives away
Active construction leaves a distinctive sequence of surface signals. Excavation exposes bare soil with high radar backscatter and low coherence between SAR passes. Sentinel-1 coherence maps, computed by differencing two 6- or 12-day passes, flag this disturbance across an entire city in a single processing run. Sites where coherence drops below roughly 0.3 in otherwise stable urban fabric are candidate construction zones. That threshold is not a hard rule; vegetated areas and water bodies produce similar values, so urban masking is required before interpretation.
Once a slab is poured and scaffolding rises, the optical signature changes. WorldView-3 at 31 cm resolves individual scaffold poles and stacked building materials. SkySat at 50 cm resolves the footprint outline, the presence of a tower crane and the approximate number of storeys under construction. Counting floor slabs in repeat imagery is a practical method for estimating completion timelines, though it requires clear skies and consistent off-nadir angles to avoid parallax artefacts.
From footprint growth to subscriber count: the arithmetic and its limits
Once a new building footprint is delineated, an estimate of subscriber yield requires assumptions about unit density and household size. Residential towers in a given city type carry reasonably consistent floor-plate-to-unit ratios; a 1,000 m² floor plate in a mid-rise block in a West African secondary city implies a different unit count than the same plate in a high-rise in a Southeast Asian capital. Operators with local market data can calibrate these ratios against their own existing subscriber density figures. Without that calibration, uncertainty on unit count runs to plus or minus 30 to 40 per cent.
Activation timing adds a further variable. Buildings in rapidly urbanising markets are often occupied floor by floor before completion. A more conservative planning assumption is that 60 to 70 per cent of units generate SIM insertions within three months of structural completion, with the remainder trickling in over the following six months. These figures are illustrative; operators should validate them against their own churn and activation records.
Google Open Buildings provides a useful baseline footprint layer for many of the regions where this problem is sharpest. Differencing a current high-resolution image against that baseline identifies net new footprints. The dataset's known limitation is that it was generated from imagery of varying vintage and does not update continuously, so it should be treated as a starting point rather than a live reference.
Combining SAR screening with optical confirmation
Running Sentinel-1 coherence loss across an entire urban area every 6 to 12 days is cheap and covers ground that no amount of commercial tasking budget would make economically viable to image at 50 cm. The practical workflow is to use coherence loss as a spatial filter: any pixel cluster above a minimum area threshold (typically 500 m² to avoid noise from road works and small demolitions) that sustains low coherence across two or more consecutive passes is flagged as a probable construction site.
Those flagged locations are then prioritised for optical tasking. A SkySat collect costs a fraction of a WorldView-3 collect and is adequate for site confirmation and footprint delineation. WorldView-3 is reserved for sites where SWIR band analysis is needed to distinguish building materials or where sub-metre structural detail matters for unit-count estimation. This tiered approach keeps tasking costs proportionate to the intelligence value of each site.
Honest constraints: cloud, resolution floors and the informal-construction problem
Cloud cover is the dominant operational constraint on optical methods. Tropical secondary cities, particularly in West Africa and maritime Southeast Asia, can sustain cloud cover exceeding 70 per cent of days during monsoon seasons. SAR coherence remains available through cloud, but it cannot resolve individual structures or count floors. In persistently cloudy conditions, the construction timeline estimate degrades to a coarser bracket: active disturbance detected, completion date unknown.
Informal construction, typically single-storey incremental builds in peri-urban areas, often falls below the spatial scale at which footprint counting is reliable. A 30 m² addition to an existing structure is detectable in WorldView-3 imagery but not in Sentinel-1 at 10 m. These additions do contribute to aggregate demand but are better captured by population-distribution methods than by construction monitoring. This page does not cover that approach.
Satellize runs the SAR coherence screening and optical change-detection pipeline on client-defined areas of interest, producing ranked site lists with estimated unit counts and confidence intervals. The Tonga crop-estimation programme demonstrated that the same change-detection architecture scales to small-area, high-cadence monitoring, though the signal physics differ.
What the output looks like in practice
The deliverable for a network planning team is a GIS layer, updated on a cadence matched to the Sentinel-1 repeat, showing active construction sites ranked by estimated gross floor area and projected completion quarter. Each site carries a confidence score reflecting the number of SAR passes over which coherence loss was sustained, whether optical confirmation has been obtained, and the quality of the footprint delineation.
That layer feeds directly into a capacity planning model. Sites projected to complete in the next two quarters trigger a review of the serving cell's headroom. Sites in the following two quarters enter a watch list. The output is not a demand forecast in itself; it is a spatially explicit lead indicator that replaces the blank space where planning-register data should be.
Typical figures
| Optical resolution (SkySat) | 50 cm pan-sharpened |
| Optical resolution (WorldView-3) | 31 cm panchromatic, 1.24 m multispectral, 3.7 m SWIR |
| SAR resolution (Sentinel-1 IW mode) | 10 m (range) × 10 m (azimuth) after multi-looking |
| Sentinel-1 revisit (equatorial) | 6 days (two-satellite constellation); 12 days (single satellite) |
| Optical tasking latency | 24 hours typical for SkySat; 1 to 4.5 days for WorldView-3 depending on off-nadir scheduling |
| Minimum detectable construction site (SAR coherence) | Approximately 500 m² sustained disturbance across two or more passes; smaller works produce noise-level signals |
| Archive depth | Sentinel-1: from 2014; WorldView-3: from 2014; SkySat: from approximately 2017 |
| Spectral bands used | Visible and NIR (optical change detection); SWIR (WorldView-3 material discrimination); C-band 5.4 GHz (SAR coherence) |
| Delivery format | GeoJSON or GeoPackage site polygons with attribute table; optional raster coherence difference layers (GeoTIFF) |
| Unit-count estimation uncertainty | ±30 to 40% without operator-supplied local calibration data; improves with validated floor-plate-to-unit ratios |
Analytics Satellize can run
| Urban construction site detection layer | SAR coherence differencing (Sentinel-1 IW SLC pairs); persistent low-coherence cluster extraction with urban mask applied | Polygon GIS layer of active construction sites, updated every 6 to 12 days, with site area and first-detection date |
| Optical site confirmation and footprint delineation | Object-based image analysis on SkySat or WorldView-3 imagery; manual QA for ambiguous sites | Confirmed building footprint polygons with floor-count estimate and structural-stage classification (foundation, frame, envelope, fit-out) |
| Estimated unit count per site | Floor-plate area × operator-supplied or regional-default unit-density ratio; sensitivity range reported | Attribute table column on site polygon layer; confidence interval expressed as low/mid/high scenario |
| Projected completion quarter | Floor-slab progression rate from repeat optical imagery; linear extrapolation with uncertainty bounds | Completion-quarter field in site layer; watch-list flag for sites completing within two quarters |
| Baseline footprint delta against Google Open Buildings | Automated footprint differencing between current delineation and published Open Buildings polygons | Net-new-footprint count and total gross floor area added since dataset vintage, per planning zone |
| Capacity headroom alert | Estimated new subscriber yield (unit count × household-size assumption × SIM penetration rate) compared against operator-supplied cell headroom thresholds | Ranked alert list of cells where projected demand increment exceeds defined headroom threshold within the planning horizon |
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.