Sensors
- Sentinel-1 SAR (ESA): C-band SAR at 5.6 cm wavelength; Interferometric Wide Swath mode delivers 10 m ground range resolution with 6-day repeat at mid-latitudes (12-day for a single satellite). Coherence between successive passes collapses when surface texture changes, making excavation and earthworks detectable even under cloud or at night.
- Planet SuperDove: 3 m optical resolution, 8 spectral bands (coastal blue to near-infrared), daily revisit over most land areas. Bare-soil exposure shows clearly in band-ratio indices. Useful for rapid confirmation after a SAR coherence alert, though cloud cover can delay optical confirmation by days to weeks depending on climate zone.
- Maxar WorldView-3: 30 cm panchromatic, 1.24 m multispectral, 16 spectral bands including SWIR. The highest commercially available optical resolution for this application; capable of resolving individual foundation trenches and distinguishing stockpile materials by spectral signature. Tasked on demand; not a persistent monitor.
- Capella Space SAR (X-band): X-band SAR at approximately 25 cm wavelength in spotlight mode, delivering sub-metre resolution. More sensitive than C-band to fine surface texture changes such as compacted gravel or freshly poured concrete. Tasking latency of hours to a day; cloud-transparent.
What bare earth gives away
The moment topsoil is stripped, three things change simultaneously: surface roughness increases, soil moisture profile alters, and the spectral reflectance shifts sharply away from vegetation. Each of those changes is measurable from orbit. A SAR sensor records backscatter intensity and, more usefully, the phase coherence between consecutive passes. Undisturbed ground maintains high coherence. Disturbed ground, whether by excavation, grading or material dumping, decorrelates the phase signal. On Sentinel-1 at 10 m resolution, coherence loss over a footprint as small as 400 to 500 square metres can be statistically significant, though reliable detection in practice sits closer to 1,000 square metres once speckle filtering is applied.
Optical sensors read the same event differently. Bare soil has a characteristic high reflectance in the red and SWIR bands and low reflectance in the near-infrared, producing a sharply negative Normalised Difference Vegetation Index. On a 3 m SuperDove image, a newly cleared plot stands out against surrounding land cover within days of stripping. The two sensor types are complementary: SAR fires through cloud and dark, optical gives spectral confirmation and higher spatial detail when skies cooperate.
Coherence change on Sentinel-1: the method in plain terms
Sentinel-1 coherence change detection compares the complex SAR phase between two co-registered passes over the same area. Where the ground has not moved, the phase relationship is stable and coherence is high (values approaching 1.0). Where something has changed, coherence drops toward zero. The technique was formalised in published literature through the 1990s and is now operationally used in the Copernicus Emergency Management Service for disaster mapping, which gives it a well-documented public track record.
For construction monitoring, the workflow is straightforward: compute coherence for a baseline pair before the period of interest, then compute coherence for each subsequent pair. Pixels that drop below a threshold (typically 0.3 to 0.4 in vegetated areas, calibrated per land cover class) and remain low across two or more consecutive pairs are flagged as changed. A single-pair drop can be noise; persistence across pairs is the signal. The 6-day Sentinel-1 revisit at mid-latitudes means a confirmed alert can arrive within 12 to 18 days of first ground disturbance. In tropical regions where Sentinel-1 coverage is thinner, revisit may extend to 12 days per pass, lengthening the detection window.
Pixel-differencing on VHR optical: faster confirmation, harder to automate
Pixel-differencing on commercial VHR imagery compares co-registered images from two dates and flags pixels whose reflectance has shifted beyond a threshold. At 30 cm (WorldView-3 panchromatic), individual foundation trenches, concrete pours and material stockpiles are resolvable. That resolution is the ceiling of what is commercially available today, and it comes with a cost: tasking fees, cloud risk, and the fact that no VHR satellite offers daily revisit over a specific parcel. WorldView-3 revisit over a given point is typically 1 to 4.5 days depending on latitude and tasking priority, but a cloud-contaminated pass yields nothing.
Automation at VHR resolution is also harder than it sounds. At 30 cm, shadows, seasonal vegetation change and parked vehicles all produce false positives in a naive differencing pipeline. Production-grade systems apply object-based image analysis and machine-learning classifiers trained on labelled construction imagery to suppress clutter. Detection recall in peer-reviewed studies on similar pipelines typically sits in the 70 to 85 percent range for sites above roughly 500 square metres, with precision dependent heavily on the false-positive tolerance set by the operator.
The limits you should plan around
Minimum detectable footprint is the first honest constraint. Sentinel-1 coherence change at 10 m resolution cannot reliably flag a single-house plot of 200 square metres; the signal is there in principle but swamped by speckle. Commercial X-band SAR from Capella at sub-metre spotlight resolution pushes the floor down considerably, but tasking cost makes it impractical as a blanket monitor over large areas. The practical architecture is tiered: Sentinel-1 as a persistent wide-area screen, commercial tasking triggered only on flagged parcels.
Cloud cover is the second constraint, and it is not trivial in temperate or tropical climates. A site in the Scottish Highlands or coastal West Africa may be cloud-obscured for weeks at a time, making optical confirmation slow. SAR sidesteps cloud but cannot distinguish a construction start from a field being ploughed unless land-use context is applied. Combining SAR change alerts with cadastral parcel boundaries and existing land-use classification reduces false positives substantially, but the classification itself carries its own error rate.
Finally, there is the archive depth question. Sentinel-1 data is freely available from 2014 onward via the Copernicus Data Space Ecosystem. Planet's archive extends to 2016 for SuperDove-class imagery. For a parcel with no recent baseline image, the first detection epoch may not arrive until several passes have been collected post-tasking.
Where this fits in a property intelligence workflow
Construction start detection is most valuable when it arrives before public information does. Planning applications in England, for example, are typically registered weeks to months after site preparation begins. A coherence-change alert on a flagged parcel can give a lender, developer or local authority a material lead time. The same signal feeds development pipeline models: aggregate construction starts across a local authority boundary, and you have a leading indicator of housing supply that no planning portal publishes in real time.
Satellize runs this kind of tiered detection pipeline on open Sentinel-1 data, with commercial tasking added on client licence for parcel-level confirmation. The Tonga crop-estimation programme demonstrated the same underlying principle at national scale: open satellite data as the persistent screen, targeted tasking for ground-truth. The architecture transfers directly to urban construction monitoring.
Typical figures
| SAR spatial resolution (Sentinel-1 IW mode) | 10 m ground range x 10 m azimuth (after multi-looking) |
| SAR spatial resolution (Capella spotlight) | Sub-metre (0.5 m in highest-resolution spotlight products) |
| Optical spatial resolution | 3 m (Planet SuperDove); 30 cm pan / 1.24 m MS (WorldView-3) |
| Revisit (Sentinel-1, mid-latitudes) | 6 days (two-satellite constellation); up to 12 days in some tropical regions |
| Revisit (Planet SuperDove) | Daily over most land areas |
| Minimum detectable disturbed footprint | ~1,000 m² reliable on Sentinel-1 coherence; ~500 m² on VHR optical with ML classifier |
| Cloud transparency | SAR (Sentinel-1, Capella): fully cloud-transparent. Optical: cloud-dependent; no useful data through opaque cloud |
| Archive depth | Sentinel-1: 2014 to present (Copernicus Data Space). Planet: 2016 to present. WorldView-3: 2014 to present (commercial archive) |
| Coherence change detection latency | 12 to 18 days from ground disturbance to confirmed alert at mid-latitudes on Sentinel-1 |
| Delivery formats | GeoTIFF change masks, GeoJSON parcel-level alerts, vector overlays compatible with GIS platforms |
Analytics Satellize can run
| Coherence-loss alert layer | Sentinel-1 interferometric coherence change detection; consecutive-pair persistence filter to suppress noise | GeoJSON alert feed of flagged parcels, updated per Sentinel-1 pass cycle |
| Bare-earth exposure map | NDVI and bare-soil index differencing on Planet SuperDove time series; threshold classification per land-cover baseline | GeoTIFF raster and polygon layer showing newly exposed soil, timestamped to image date |
| Parcel-level construction start report | Fusion of SAR coherence alert and optical bare-earth confirmation; cross-referenced against cadastral boundary and land-use classification | PDF or structured JSON report per flagged parcel, including first-detection date, confidence score and imagery thumbnails |
| Area-wide construction start index | Aggregation of parcel-level alerts within user-defined administrative or market boundaries; time-series smoothing to produce a leading indicator | Monthly time-series chart and tabular data export by local authority or custom polygon |
| VHR confirmation image and annotation | Tasked WorldView-3 or Capella acquisition on triggered parcels; object-based image analysis to classify excavation, stockpile and hardstanding features | Annotated GeoTIFF with feature polygons and classification labels; delivered within 24 to 72 hours of tasking |
| Pre-application activity flag | Comparison of satellite-detected start date against public planning register; gap analysis to quantify lead time over official data | Structured table of parcels with satellite-detected activity preceding planning registration, with date-gap column |
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.