Modular and volumetric building factory output tracking
Offsite construction yards hold the truth about factory output and project delivery. Very-high-resolution tasked imagery from Pléiades Neo and SkySat, analysed with object-based methods, can count and classify volumetric modules before a single delivery note is signed.
Sensors
- Pléiades Neo (Airbus): 30 cm panchromatic, 50 cm multispectral (4-band plus red-edge and deep blue). Constellation of four satellites gives up to four looks per day over a tasked site. At 30 cm, a standard 3.6 m wide volumetric module subtends roughly 12 pixels across its short axis, which is enough to distinguish module type by footprint but not to read markings.
- SkySat (Planet): 50 cm panchromatic, ~1 m multispectral. Fleet of 21 satellites supports multiple tasking passes per day in many mid-latitude locations. Slightly coarser than Pléiades Neo but sufficient for module counting and gross footprint classification; stack-height inference from shadow length is reliable at this resolution.
- Maxar WorldView Legion: 30 cm panchromatic. Six-satellite constellation designed for high-revisit tasking over priority areas. Stereo and tri-stereo collection modes allow photogrammetric height estimation of stacked modules, with published height accuracy in the 20–50 cm range under good geometry, which is marginal for distinguishing a single-module stack from a double.
- WorldView-3 (Maxar, archive): 31 cm panchromatic, 1.24 m multispectral, plus SWIR bands. Archive depth back to 2014 supports retrospective baseline construction for established factories. SWIR bands help separate galvanised steel module skins from concrete raw-material stockpiles, which is useful for object classification.
What a yard full of modules actually tells you
Offsite construction factories operate on a pull schedule: modules are manufactured against a project programme, staged in yard storage, and despatched to site in sequence. Yard inventory at any moment is therefore a direct function of factory throughput, site readiness and logistics capacity. When modules accumulate beyond planned buffer stock, either the factory is running ahead of site demand or deliveries are being held. When the yard empties faster than expected, the factory may be struggling to keep pace. Neither condition is visible in a contractor's progress report until it becomes a formal delay.
Satellite imagery makes the yard legible on a near-daily basis without any cooperation from the factory operator. That is the core value: an independent, timestamped count of physical objects in a defined geographic boundary, updated as often as tasking allows.
Resolution floors and what they actually resolve
A standard volumetric module for residential construction is typically 3.5 to 4.5 metres wide and 8 to 14 metres long. At Pléiades Neo's 30 cm panchromatic resolution, the short axis of a 3.6 m module covers about 12 pixels. That is enough to detect the module, estimate its footprint class and, using shadow geometry, infer whether it is stacked one or two high. It is not enough to read identification markings or distinguish finish state from roof texture alone.
Single-module movements between consecutive passes are detectable only if the revisit interval is short enough and the module has moved far enough to exit its previous pixel cluster. In practice, a module shifted by one bay width (roughly 5–6 m) will register as a change at 30 cm resolution. A module rotated in place within the same bay will not. This is an honest limit: satellite change detection tracks inventory level and spatial redistribution, not individual unit provenance.
Stack height estimation from shadow length is geometry-dependent. A sun elevation below about 30 degrees produces long shadows that are easy to measure; above 60 degrees, shadows compress and height discrimination between a single-stacked and double-stacked module becomes unreliable. Analysts should record solar geometry at acquisition time and flag collections where elevation exceeds 55 degrees as having degraded height confidence.
Object-based analysis: separating modules from everything else in the yard
Pixel-by-pixel classification fails in a modular factory yard because the spectral signature of a galvanised steel module roof is similar to that of a corrugated storage shed, a flatbed trailer or a raw-steel coil bay. Object-based image analysis (OBIA) addresses this by segmenting the image into coherent objects first, then classifying by a combination of spectral mean, shape index, size, orientation and context.
Modules have a characteristic aspect ratio (length-to-width typically 2.5:1 to 4:1), a regular rectangular geometry, and they appear in parallel rows with consistent inter-object spacing. Raw-material bays tend to be larger, less regular in outline, and spectrally heterogeneous across their surface. Vehicles are smaller and oriented randomly. OBIA rule sets built on these geometric and contextual features can achieve module detection rates that manual digitising studies on WorldView imagery have demonstrated at above 90 percent for well-separated stacks, dropping to around 70–80 percent when modules are packed tightly and shadows overlap. Those figures come from published remote-sensing literature on industrial object detection, not from Satellize's own trials.
Multispectral bands add a useful separation layer. WorldView-3's SWIR bands (1195–2365 nm range, eight bands) respond differently to galvanised steel, painted steel, concrete and timber, which are the dominant surface materials in a mixed yard. Where only four-band visible-to-NIR imagery is available, the separation relies more heavily on shape and context, and classification confidence is lower for ambiguous objects near yard boundaries.
Revisit cadence and the latency problem
The commercial very-high-resolution constellations available today can, in principle, task a specific yard multiple times per day. In practice, cloud cover, competing tasking demand and satellite geometry reduce effective revisit. A realistic planning assumption for a mid-latitude site in northern Europe or North America is two to four clear acquisitions per week under average cloud climatology, with periods of several consecutive clouded days that produce data gaps.
For project-finance or contractual monitoring purposes, a weekly cadence is generally sufficient to track inventory trends. For operational logistics, where a client needs to know whether a delivery batch has left the yard in the past 24 hours, cloud gaps are a genuine operational risk. Synthetic aperture radar (SAR) can partially fill cloud gaps for change detection, but at the resolutions currently available from commercial SAR constellations, individual module discrimination is not reliable. SAR is useful for detecting whether the yard boundary has changed in extent, not for counting modules within it. This page does not cover SAR methods in depth; that ground is held by the precast concrete yard inventory monitoring page in this library.
Latency from acquisition to delivered analysis is a separate constraint. Tasked imagery from Pléiades Neo and SkySat is typically available for download within two to six hours of acquisition. Processing through an OBIA pipeline adds time depending on automation level. A fully automated pipeline can deliver a module count update within the same business day; a manually reviewed output takes one to two days. Clients should specify which latency tier they require before setting tasking schedules.
From pixel counts to programme intelligence
Raw module counts become useful only when placed against a baseline. The baseline can be a contractually agreed production schedule (how many modules should be in the yard on a given date), a historical inventory curve from archive imagery, or a peer-factory comparison drawn from open-source imagery of comparable sites. The choice of baseline determines what the analysis can actually assert.
Against a contractual schedule, the analysis supports covenant monitoring: a lender or equity investor can verify that factory throughput is consistent with the drawdown milestones in a development finance agreement. Against a historical curve, the analysis supports early-warning flagging: a yard that is running two standard deviations above its typical pre-delivery inventory level is worth a phone call before it becomes a programme delay. Against peer factories, the analysis supports competitive intelligence: understanding whether a rival manufacturer's yard is growing or shrinking is a legitimate input to procurement strategy.
Satellize structures analytics engagements around client-defined programme milestones rather than fixed reporting calendars. The Tonga crop-estimation programme showed that periodic-update analytics tied to agricultural cycles outperform continuous feeds for clients who need actionable signals rather than data volume. The same principle applies here: a module-count alert triggered when inventory crosses a defined threshold is more useful than a daily report that requires human triage.
Honest limits and what to do about them
Three limits deserve plain statement. First, satellite imagery cannot distinguish a finished module from an unfinished one unless the exterior cladding state is visually distinct at 30–50 cm resolution. Interior fit-out is invisible. A yard count is a count of physical objects, not a quality-assured delivery count.
Second, modules moved inside a factory building are invisible to optical sensors. If a factory operator stages finished modules under a covered canopy before moving them to the open yard, the satellite record will undercount production. Ground truth from delivery manifests or periodic site visits remains necessary to calibrate the optical inventory signal.
Third, the method works best for factories with open, hard-standing yards and consistent module geometry. Factories that store modules under temporary fabric structures, or that produce highly variable module sizes, require bespoke classification rule sets and will have higher uncertainty bounds. Buyers should ask for site-specific validation before committing to a monitoring programme.
Typical figures
| Best available spatial resolution | 30 cm panchromatic (Pléiades Neo, WorldView Legion) |
| Multispectral resolution | 50 cm (Pléiades Neo); 1.24 m (WorldView-3) |
| Effective tasked revisit (clear sky) | 2–4 acquisitions per week at mid-latitudes under average cloud; up to 4 per day in principle |
| Minimum detectable module footprint | ~3.5 m × 8 m (well-separated, open yard); smaller or touching modules have higher miss rate |
| Stack height discrimination | Single vs. double stack detectable when solar elevation is below ~55°; unreliable above that |
| Spectral bands | Panchromatic + 4-band VNIR standard; WorldView-3 adds 8-band SWIR for material separation |
| Archive depth | WorldView-2/3 archive from 2009/2014; Pléiades from 2012; SkySat from ~2016 |
| Analysis latency (automated pipeline) | Same business day from acquisition; manually reviewed output 1–2 days |
| Delivery formats | GeoJSON module polygons, GeoTIFF annotated imagery, CSV inventory time series, PDF milestone report |
| Cloud-gap risk | Multi-day gaps likely in maritime climates; SAR supplements for perimeter change only, not module count |
Analytics Satellize can run
| Module inventory count | Object-based image analysis (OBIA) on tasked panchromatic imagery; geometric rule set (aspect ratio, size, row spacing) | GeoJSON polygon layer with module count per zone; CSV time-series updated per acquisition |
| Stack height classification | Shadow-length photogrammetry using solar azimuth and elevation at acquisition time | Per-object height class attribute (single/double) appended to inventory GeoJSON; flagged where solar geometry degrades confidence |
| Footprint-class discrimination | OBIA shape index combined with multispectral band ratios (SWIR where available) to separate module types from raw-material bays and vehicles | Classified inventory layer with module-type labels and confidence scores |
| Inventory trend and threshold alert | Time-series analysis of per-acquisition module counts against client-supplied programme schedule or historical baseline | Automated alert (email or API webhook) when inventory crosses defined upper or lower threshold |
| Change-detection delta map | Image differencing and object-level change attribution between consecutive acquisitions | GeoTIFF and GeoJSON showing modules added, removed or relocated between two dates |
| Covenant compliance report | Inventory count compared to contractual milestone table; narrative interpretation of variance | PDF report suitable for lender or equity investor review at defined programme milestones |
| Retrospective baseline construction | Archive imagery analysis (WorldView-2/3, Pléiades) to reconstruct historical inventory curve for an established factory | CSV and chart of monthly inventory levels over archive period; used to calibrate alert thresholds |
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.