Development pipeline tracking across urban regions

Multi-temporal optical and SAR imagery can identify active construction sites weeks or months before they appear in official permit registers, giving investors and planners a forward view of the urban development pipeline.

Sensors

Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit at mid-latitudes with both satellites. Ideal for area-wide monitoring of bare-earth exposure, vegetation removal and impervious surface gain across entire metropolitan regions.
Sentinel-1 SAR (C-band): Interferometric coherence time-series at 20 m ground range resolution, 6-day revisit per satellite. Coherence loss reliably flags ground disturbance and active earthworks; coherence recovery signals structural completion. Cloud-independent, which matters in wet climates.
Planet SkySat: 50 cm resolution optical imagery, taskable to specific sites. Resolves individual foundation forms, structural frames and roofing materials. Revisit is on-demand rather than systematic, so it is best used to confirm phase transitions flagged by Sentinel.
Maxar WorldView Legion: 30 cm panchromatic resolution, up to 15 revisits per day over target cities once the full constellation is operational. At this resolution, construction equipment, material stockpiles and partial roof coverage are individually distinguishable, enabling fine-grained phase classification.

What a permit database cannot tell you

Official planning registers record intent, not action. A developer submits drawings, receives consent and then does nothing for six months, or breaks ground quietly before consent is formally logged. In many jurisdictions, the gap between first earthwork and first permit entry is routinely four to twelve weeks. In markets with lighter enforcement, the gap is longer.

Satellite imagery observes the ground, not the paperwork. Bare-earth exposure, the spectral signature of disturbed soil in Sentinel-2's red-edge and SWIR bands, is detectable from the first machine pass. That signal is independent of what any authority has or has not recorded. For anyone pricing land, underwriting construction finance or planning infrastructure capacity, that timing advantage is the core value proposition.

How optical time-stacks classify construction phase

A multi-temporal optical stack is simply a ranked sequence of images over the same footprint. Change between epochs is computed per pixel. The spectral trajectory of a construction site is reasonably consistent: vegetated or sealed surface gives way to bare soil, bare soil acquires the higher reflectance of concrete or pale aggregate, and eventually the near-infrared response of a completed roof stabilises. Sentinel-2's 10 m pixels resolve this trajectory at the scale of individual plots in low-density suburban development, though a single 10 m pixel can straddle a plot boundary in dense urban infill, which introduces classification ambiguity.

Normalised Difference Built-up Index (NDBI) and Bare Soil Index (BSI) computed from Sentinel-2 bands B11 and B8A are well-documented in the published literature for exactly this purpose. The practical limit is that a 5-day revisit means a phase transition that completes in fewer than five days may be missed or appear to jump a stage. Cloud cover compounds this: in tropical or maritime climates, cloud-free observations may arrive only every two to four weeks, collapsing the precision of phase-transition dating considerably.

SAR coherence adds what cloud takes away

Synthetic aperture radar penetrates cloud. Sentinel-1's C-band coherence time-series has become a standard method for construction monitoring precisely because it does not depend on optical conditions. Coherence between two SAR acquisitions six days apart is high over stable surfaces and collapses over surfaces that have moved or changed. Active earthworks, concrete pours and structural erection all produce coherence loss. Once a building envelope is complete and the surface stabilises, coherence recovers.

The limitation is spatial. At 20 m ground range resolution, Sentinel-1 cannot resolve individual structures in dense urban blocks; it resolves clusters. For a city-scale pipeline survey, that is often sufficient. For a specific contested site, a commercial SAR asset at finer resolution is needed. ICEYE and Capella Space operate X-band SAR at sub-metre resolution, though their data sits behind commercial licences and is not freely available the way Sentinel-1 is.

Fusing optical and SAR coherence time-series improves phase-classification accuracy compared with either source alone. The two datasets fail under different conditions, so their combination is more reliable across a full calendar year than either individually.

The revisit problem and what it means for pipeline precision

A development pipeline product is only as precise as the observation frequency that underlies it. If the goal is to report that a site broke ground this week, a 5-day optical revisit with typical cloud contamination may deliver a ground-break date accurate to plus or minus two weeks. That is useful for strategic market intelligence. It is not sufficient for, say, a construction lender trying to verify a drawdown milestone against a specific completion date.

Higher revisit from commercial constellations narrows the uncertainty. Planet's Dove constellation provides daily optical coverage at 3 to 5 m resolution, which is coarser than SkySat but systematic. At 3 m, rooftop completion on a standard residential unit is detectable. Daily revisit reduces phase-transition dating error to a few days under clear-sky conditions. The cost is that daily commercial coverage over a large metropolitan region is not cheap, and the economics only work if the pipeline intelligence is priced accordingly.

The honest framing for any buyer: Sentinel-based pipelines are cost-effective for regional surveys and strategic planning horizons. VHR commercial tasking is cost-effective for specific sites where the financial decision depends on precise timing.

From pixels to a pipeline product

The output that matters to a real estate investor or a planning authority is not a stack of images; it is a structured list of sites, each with a geographic footprint, a detected start date, a current phase classification and an estimated completion window. That list should be updated on each new observation cycle and should flag sites that have stalled, defined as no phase progression over a defined period, which is itself a signal of financial or regulatory distress.

Phase classification is typically a supervised model trained on labelled examples: bare earth, foundation, frame, enclosed shell, completed roof. Each class has a spectral or coherence signature. The model assigns a probability to each class at each epoch. Transitions are logged when the dominant class changes. Stall detection is a secondary rule applied when the dominant class is unchanged for a user-defined number of consecutive observations.

Satellize runs this kind of multi-temporal classification on open Sentinel data, with commercial VHR tasking added for specific sites on client licence. The Tonga crop-estimation programme uses a similar temporal-stack methodology, adapted to agricultural phenology rather than construction phase, which illustrates that the underlying approach transfers across domains.

Honest limits before you commission anything

Cloud is the most obvious constraint and is often underestimated by buyers in temperate or tropical markets. A site in coastal British Columbia or monsoon-season Mumbai may yield fewer than eight cloud-free Sentinel-2 observations per quarter. SAR partially compensates, but SAR coherence is noisier in vegetated or sloped terrain.

Dense urban cores present a second problem: building shadows and geometric layover in SAR obscure ground-level activity between tall structures. A basement excavation surrounded by existing buildings may be invisible from orbit. Ground-break on a tower replacement site in a city centre is harder to detect than ground-break on a greenfield suburban estate.

Phase classification accuracy in published studies using Sentinel-2 and Sentinel-1 fusion typically falls in the range of 80 to 90 percent at the site level, with the main confusion between the foundation and early-frame stages. Buyers should treat phase assignments as probabilistic indicators, not legal records. For compliance or contractual purposes, satellite-derived phase assessments need ground verification.

Typical figures

Area-wide optical resolution	10 m (Sentinel-2 MSI, visible and NIR bands)
VHR optical resolution	30 cm panchromatic (Maxar WorldView Legion); 50 cm (Planet SkySat)
SAR resolution	20 m ground range (Sentinel-1 IW mode); sub-metre available from commercial X-band assets
Sentinel-2 revisit	5 days at mid-latitudes (both satellites combined); effective cloud-free revisit varies by climate zone
Sentinel-1 coherence revisit	6 days per satellite; 12-day coherence pairs standard
Minimum detectable site area	Approximately 400 m² (2 × 2 Sentinel-2 pixels); smaller sites require VHR confirmation
Phase-transition dating precision	Plus or minus 5 to 14 days (Sentinel optical, clear sky); plus or minus 2 to 5 days (daily VHR, clear sky)
Archive depth	Sentinel-2 from 2015; Sentinel-1 from 2014; Landsat back to 1972 for historical baseline
Key spectral inputs	SWIR (B11, B12), red-edge (B05, B06), NIR (B08) for NDBI and BSI; C-band backscatter and coherence for SAR
Delivery formats	GeoJSON site polygons, GeoTIFF change layers, CSV pipeline register, dashboard feed

Analytics Satellize can run

Ground-break detection alert	Bare Soil Index change threshold on Sentinel-2 multi-temporal stack, confirmed by Sentinel-1 coherence loss	Automated alert with site polygon, detected date and confidence score, delivered as GeoJSON or API push
Construction phase classification layer	Supervised pixel classification (random forest or gradient boosting) trained on labelled optical and SAR features per phase class	Raster and vector layer updated each observation cycle, with dominant phase label and class probabilities per site
Pipeline register with phase history	Temporal state machine tracking dominant phase transitions across the observation archive	Structured CSV or database table: site ID, footprint area, detected start date, current phase, last observation date, phase-transition log
Stalled-site flag	No phase progression detected over a configurable observation window (typically 60 to 90 days), applied as a secondary rule on the phase time-series	Flagged subset of the pipeline register, suitable for credit-risk or distressed-asset screening
Roof-on completion estimate	Coherence recovery plus stabilisation of NDBI signal, cross-checked against VHR imagery where tasked	Completion-date estimate with confidence interval, appended to pipeline register
Regional pipeline summary report	Aggregation of site-level detections by administrative boundary, sector type and phase distribution	Quarterly PDF and machine-readable summary: active site count, new starts, completions, stalls, by district

Sources

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.