Sensors
- Pléiades Neo (Airbus): 30 cm panchromatic, 50 cm multispectral, same-day revisit with constellation of four satellites. Shadow length at known sun elevation gives stack height to roughly ±0.3 m. The primary optical sensor for individual segment discrimination.
- Capella Space X-band SAR: Spotlight mode delivers 50 cm resolution; Sliding Spotlight 1 m. X-band (9.65 GHz) backscatter is sensitive to the flat, parallel concrete surfaces of stacked segments, which produce strong double-bounce returns distinguishable from soil or gravel. Operates through cloud and at night. Revisit typically 12–24 hours to any point globally with tasking.
- SPOT 7 (Airbus): 1.5 m panchromatic, 6 m multispectral. Insufficient resolution to count individual segments but adequate for tracking gross yard occupancy and change in stacked-area footprint between epochs. Useful as a lower-cost monitoring layer between high-resolution collections.
- Planet SuperDove: 3 m multispectral, daily revisit. Cannot resolve individual precast units but provides a daily change signal for yard footprint and vehicle presence, flagging dates when a high-resolution collection is worth tasking.
What a shadow tells you that a pixel cannot
A precast tunnel segment stack imaged at 30 cm resolution appears as a pale rectangular object. Without additional information, it could be a shipping container, a concrete barrier, or a pile of formwork. Shadow geometry resolves the ambiguity. At a known solar elevation angle, the shadow cast by a stack on flat hardstanding gives its height directly: height equals shadow length multiplied by the tangent of the solar elevation. For Pléiades Neo imagery collected at a solar elevation of 45 degrees, a shadow 1.2 m long implies a stack 1.2 m tall. Standard tunnel lining segments for a 6 m bore typically run 300–400 mm thick per ring element; a 1.2 m shadow therefore suggests a stack of three to four rings.
The method has real limits. It requires unobstructed shadow fall on a flat, contrasting surface. Yards with multiple adjacent stacks produce overlapping shadows that defeat simple geometric inversion. Overcast conditions reduce shadow contrast. And the approach assumes the analyst knows the segment geometry in advance, which is usually available from project documentation but not always from satellite data alone. Where shadows are clean, counting accuracy on individual stacks in controlled trials with VHR imagery has been reported in the published literature at better than 90 percent for objects larger than roughly 1 m in their smallest dimension.
Why X-band radar sees concrete differently from everything else in the yard
SAR backscatter intensity depends on surface roughness relative to wavelength and on the geometry of scattering structures. X-band wavelength is approximately 3.1 cm. Smooth concrete surfaces are specular at X-band and return little energy back to the sensor when viewed at typical incidence angles of 25–50 degrees. But a stack of flat segments sitting on a concrete hardstanding creates a dihedral corner reflector between the vertical face of the stack and the horizontal ground plane. Dihedral returns are very bright in SAR imagery, often 10–15 dB above surrounding clutter, and they are geometrically stable: the bright return appears at a predictable offset from the true stack position due to SAR layover.
This is what separates a segment stack from a gravel stockpile or a soil berm. Gravel produces diffuse volume scattering, relatively uniform and moderate in intensity. A concrete stack produces a sharp, intense dihedral line. Change detection between two Capella Spotlight collections separated by days or weeks will show new bright returns where stacks have been added and the disappearance of returns where stacks have been removed. The practical detection floor is roughly a stack with a vertical face of at least 0.5–0.8 m height and 1 m width at Capella's 50 cm Spotlight resolution, though yard clutter from vehicles, fencing and other metal structures introduces false positives that require filtering.
Fusing optical and SAR: where each sensor earns its keep
Optical and SAR detections are complementary rather than redundant. Optical imagery under clear skies gives the cleaner count: segment outlines are visible, colour contrast between concrete and hardstanding aids segmentation, and shadow geometry adds height. SAR works regardless of cloud cover and can be collected at night, which matters on active yards with 24-hour logistics. A fusion approach tasks Capella daily or every two days for change detection, then triggers a Pléiades Neo collection on days when SAR shows significant change, to get the precise count and confirm object class.
The fusion also handles a common failure mode. Large flat objects such as precast wall panels for modular buildings have a low profile and cast short shadows, making them hard to count optically. Their flat horizontal face, however, is a strong specular reflector at steep SAR incidence angles, producing a distinctive low-backscatter footprint surrounded by a bright edge return. Combining the two signatures reduces misclassification. Neither sensor alone is sufficient for all object types in a mixed precast yard.
Clutter, occlusion and the honest detection floor
Precast yards are not tidy. Active yards contain mobile cranes, flatbed trucks, temporary offices, rebar bundles, formwork stacks and fuel tanks, all of which produce SAR returns and cast shadows. Object-based image analysis pipelines trained on VHR optical imagery can achieve high recall for segment stacks, but precision degrades sharply when yard density is high and objects are partially occluded by cranes or adjacent stacks.
The minimum detectable stack in Pléiades Neo imagery is approximately 0.6 m in the smallest plan dimension, which corresponds to roughly two pixels at 30 cm resolution. Below that threshold, stacks merge with background texture. In Capella Spotlight mode, the dihedral return from a stack shorter than about 0.5 m becomes indistinguishable from ground clutter at typical signal-to-noise ratios. Stacks stored under temporary shade structures or fabric covers are not detectable by optical means and their SAR signature is attenuated. These are real constraints. A programme relying on satellite counts alone should carry an uncertainty margin of roughly 10–20 percent on total unit count in a busy, cluttered yard, narrowing to 5 percent or better in well-organised yards with clear hardstanding.
Turning inventory counts into schedule intelligence
A raw segment count becomes commercially useful when set against the project's planned consumption rate. If a tunnel drive is designed to consume 120 ring sets per week and the yard count drops by 600 units over five weeks without a corresponding delivery record, the programme is either ahead of plan or deliveries are being missed. Neither inference is trivial: the first may trigger early procurement of the next batch; the second may indicate a supply-chain problem that project finance covenants require to be reported.
Time-series inventory curves derived from bi-weekly satellite collections can be compared against planned drawdown schedules to produce a simple schedule-health index. Satellize applies this kind of throughput-rate analysis in its analytics work, including the Tonga crop-estimation programme where similar time-series change methods underpin the core workflow. The method generalises readily to precast yards, though the object classes and detection logic differ substantially from agricultural applications.
Delivery of the analytic product is typically a GIS layer of detected stack polygons with associated height estimates, a count table by yard zone, and a change report flagging anomalous drawdown or accumulation events. Update frequency is matched to the project's reporting cycle, commonly fortnightly for project-finance purposes and weekly for active construction monitoring.
Typical figures
| Best optical resolution (Pléiades Neo) | 30 cm panchromatic, 50 cm multispectral |
| Best SAR resolution (Capella Spotlight) | 50 cm (Spotlight); 1 m (Sliding Spotlight) |
| SAR frequency | X-band, approx. 9.65 GHz (Capella Space) |
| Optical revisit (Pléiades Neo constellation) | Same-day revisit possible; routine tasking 1–2 days |
| SAR revisit (Capella) | 12–24 hours to any point globally with tasking |
| Minimum detectable stack (optical) | Approx. 0.6 m smallest plan dimension at 30 cm resolution |
| Minimum detectable stack height (SAR dihedral) | Approx. 0.5–0.8 m vertical face at 50 cm SAR resolution |
| Stack height estimation accuracy (shadow geometry) | ±0.3 m under clear skies, flat hardstanding, known solar geometry |
| Archive depth | Pléiades/SPOT: from 2012; Capella: from 2020 |
| Delivery formats | GeoJSON polygon layers, GeoTIFF change rasters, CSV count tables, PDF change reports |
Analytics Satellize can run
| Segment stack count by yard zone | Object-based image analysis (OBIA) on Pléiades Neo panchromatic imagery; morphological filtering by aspect ratio and area to isolate rectangular concrete objects | GeoJSON polygon layer with unit-count attribute per stack, updated per collection |
| Stack height estimate | Solar shadow length measurement at known sun elevation angle; height = shadow length × tan(solar elevation) | Attribute field in stack polygon layer; tabular summary by zone |
| SAR backscatter change detection | Coherent or intensity change detection between Capella Spotlight pairs; dihedral return classification by intensity threshold and spatial pattern | Change raster (GeoTIFF) with added/removed stack footprints flagged; daily or bi-daily alert feed |
| Yard occupancy index | Ratio of detected stack footprint area to total hardstanding area, derived from Planet SuperDove daily imagery for trend and VHR for calibration | Time-series CSV; weekly occupancy chart in PDF report |
| Throughput rate and schedule-health index | Differencing of sequential inventory counts against client-supplied planned consumption schedule; anomaly flagging where drawdown rate deviates beyond agreed threshold | Fortnightly schedule-health report; breach alert when deviation exceeds threshold |
| Object classification confidence map | Fusion of optical OBIA detections and SAR dihedral-return detections; objects confirmed by both sensors assigned high confidence; single-sensor detections flagged for review | Classified GeoJSON with confidence attribute; analyst-reviewed summary for ambiguous objects |
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.