Construction material stockpile volume tracking from stereo imagery
Repeat stereo satellite imagery can estimate stockpile volumes of aggregate, sand and prefabricated components to within a few percent, giving supply-chain auditors and insurers an independent check without setting foot on site.
Sensors
- Airbus Pléiades Neo: 30 cm panchromatic resolution with native tri-stereo acquisition in a single pass. Three viewing angles (forward, nadir, backward) in one overpass reduces temporal parallax to seconds, which is the geometry needed to resolve steep stockpile faces accurately. Revisit approximately 1 day at mid-latitudes with the four-satellite constellation.
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral. Capable of along-track stereo and tri-stereo. The SWIR bands (eight bands, 3.7 m) can help distinguish material type by surface reflectance, which is useful when aggregate and sand piles sit adjacent. Revisit roughly 1 day at high off-nadir angles.
- Planet SkySat: 50 cm native resolution, video and still-frame tasking. Stereo pairs can be collected within a single overpass. Revisit is flexible under tasking, but the stereo baseline is shallower than Pléiades Neo, which reduces height sensitivity on low-profile stockpiles. Best suited to change detection between epochs rather than single-epoch volumetric precision.
- SPOT 6/7: 1.5 m panchromatic. Stereo and tri-stereo modes available. The coarser ground sampling distance limits usefulness to large stockpiles (typically above roughly 5 m height and tens of metres across) but the wider swath (60 km) covers entire industrial parks in one pass, useful for portfolio-level auditing.
Why a satellite can measure a pile of gravel
Photogrammetric volume estimation from satellite imagery rests on a simple principle: if you photograph the same object from two or more known angles, you can triangulate the three-dimensional position of every visible surface point. The result is a digital surface model (DSM). Subtract a pre-construction or cleared-site DSM and you have the volume above the reference plane. For aggregate stockpiles, the geometry is almost ideal: the material is opaque, the surface texture is rough enough to anchor tie-points, and the boundary between pile and ground is usually sharp.
The critical variable is the base-to-height ratio of the stereo pair. Pléiades Neo tri-stereo, with convergence angles typically between 10 and 25 degrees, achieves vertical accuracy in the range of 0.2 to 0.5 m (one standard deviation) over well-textured terrain according to published validation studies using ground control. WorldView stereo performs similarly. That translates to volume errors of roughly 2 to 5 percent on a large, well-shaped conical stockpile, rising steeply for small or irregularly shaped ones.
What a steep face hides from a nadir sensor
The honest limit of any nadir-looking or near-nadir satellite is occlusion. A stockpile with a face angle steeper than the sensor's off-nadir viewing angle will cast a shadow and hide the lower portion of that face entirely. Aggregate repose angles are typically 35 to 40 degrees. At nadir, a sensor sees only the top and gently sloping shoulders; the steep flanks appear as a compressed strip or vanish into shadow. The DSM interpolates across the gap, which introduces systematic underestimation of volume.
Tri-stereo acquisition partially addresses this by providing one forward and one backward oblique view alongside the nadir image. The oblique views can see into faces invisible to nadir, reducing the shadowed fraction. Even so, a pile immediately adjacent to a tall structure, or a concave stockpile bay enclosed on three sides, will retain blind zones. Buyers should expect residual underestimation on enclosed or very steep stockpiles and should request oblique tasking angles of 20 degrees or more when the geometry of the site is known in advance.
Tri-stereo versus stereo pair: the accuracy gap matters
A conventional stereo pair gives one disparity map. A tri-stereo acquisition gives two overlapping disparity maps that can be merged, filling occlusion gaps from one direction with the other view. Published comparisons of Pléiades tri-stereo DSMs against airborne lidar reference data over quarry and industrial sites report root-mean-square height errors of 0.3 to 0.6 m without ground control points, improving to 0.15 to 0.3 m with even a handful of GPS-surveyed ground control points placed at the site perimeter.
For supply-chain auditing, the practical question is whether the volume uncertainty is smaller than the commercial tolerance. A 3 percent volume error on a 50,000-tonne aggregate stockpile is 1,500 tonnes, which may be acceptable for insurance valuation but too coarse for daily batching-plant inventory control. Satellite stereo sits comfortably in the audit and covenant-monitoring tier. It is not a substitute for a surveyor's drone on a single critical day, but it provides independent, timestamped evidence across every repeat acquisition, which a drone survey rarely does.
Material discrimination: knowing what you are measuring
Volume alone is ambiguous if the site holds multiple material types. A DSM cannot distinguish a sand pile from a crushed-limestone pile of identical shape. WorldView-3's eight-band multispectral and eight-band SWIR imagery offers a partial answer. Published spectral libraries show that limestone, granite aggregate, sand and fly ash have distinct reflectance signatures in the SWIR, particularly around 1,600 nm and 2,200 nm. Supervised classification using these bands can label material type with reasonable confidence in clear-sky conditions, though wet surfaces and dust contamination shift spectra and reduce separability.
Pléiades Neo's four-band multispectral (blue, green, red, near-infrared at 30 cm) is less discriminating spectrally but sufficient to separate dark basalt aggregate from pale limestone or sand by simple band ratios. Prefabricated concrete components, with their regular geometry and pale grey tone, are identifiable by shape rather than spectral signature. The practical workflow combines spectral classification for material labelling with DSM differencing for volume, then assigns density assumptions per material class to convert volume to mass.
Repeat cadence and the question of when to task
A single stereo acquisition gives a snapshot. The supply-chain value comes from repeat tasking at intervals matched to the delivery and consumption rhythm of the site. A concrete batching plant on a major infrastructure project might turn over its aggregate stockpile every two to four weeks. Monthly tasking would catch the trend; weekly tasking would catch individual delivery events. Cloud cover is the practical constraint: at a site with 60 percent cloud probability, a weekly tasking request will on average yield a clear acquisition every two to three weeks.
Archive depth is an underused resource. Pléiades has collected stereo imagery over many major construction sites since 2012, and WorldView stereo archive goes back further. Retrospective volume reconstruction, month by month, can reveal historical stockpile drawdown patterns relevant to dispute resolution or project-finance covenant compliance reviews. Satellize structures repeat-tasking programmes around client reporting cycles and supplements tasked imagery with archive pulls when a historical baseline is needed.
One practical note on latency: stereo DSM production from raw imagery takes hours, not days, with modern photogrammetric pipelines. Delivery of a volume estimate within 24 hours of image acquisition is achievable. The binding constraint is cloud clearance, not processing time.
Accuracy, limits and what the number actually means
Volume estimates from satellite stereo are not survey-grade. They carry uncertainty from three sources: DSM height error (0.2 to 0.6 m depending on geometry and ground control), the choice of reference plane (a poorly defined base elevation inflates volume error significantly), and the density assumption used to convert volume to mass. Bulk density of aggregate varies from roughly 1,400 kg/m³ for loose sand to over 1,700 kg/m³ for compacted crushed stone. A 10 percent density uncertainty propagates directly into a 10 percent mass uncertainty regardless of how good the DSM is.
For insurance valuation and supply-chain auditing, the appropriate framing is not point accuracy but consistency across repeat acquisitions from the same sensor and geometry. Systematic errors cancel when you are measuring change rather than absolute volume. A site that shows a 30 percent reduction in stockpile volume between two acquisitions is informative even if the absolute volume carries a 5 percent uncertainty. Buyers who understand this distinction get the most from the method.
Typical figures
| Best available ground sampling distance | 30 cm (Pléiades Neo panchromatic), 31 cm (WorldView-3 panchromatic) |
| Typical DSM vertical accuracy (tri-stereo, GCP-assisted) | 0.15 to 0.3 m RMSE over well-textured terrain |
| Typical DSM vertical accuracy (tri-stereo, no GCP) | 0.3 to 0.6 m RMSE; degrades further over smooth or shadowed surfaces |
| Minimum stockpile detectable volumetrically | Approximately 500 m³ for Pléiades Neo tri-stereo; larger for SPOT 6/7 (1.5 m GSD) |
| Revisit under tasking | Approximately 1 day (Pléiades Neo 4-satellite constellation, WorldView-3 high off-nadir); 1 to 3 days (SkySat) |
| Stereo acquisition modes | Along-track stereo pair or tri-stereo (Pléiades Neo, WorldView-3, SPOT 6/7); single-pass stereo (SkySat) |
| Spectral bands for material classification | 4-band VNIR at 30 cm (Pléiades Neo); 8-band VNIR + 8-band SWIR at 1.24/3.7 m (WorldView-3) |
| Archive depth | Pléiades from 2012; WorldView series from 2007; SPOT 6/7 from 2012 |
| Typical delivery latency | Volume report within 24 hours of cloud-free acquisition |
| Deliverable formats | GeoTIFF DSM, point cloud (LAS), GIS polygon layer with volume attributes, PDF audit report |
Analytics Satellize can run
| Stockpile volume per material class | Semi-global matching photogrammetry on tri-stereo imagery; DSM differencing against reference terrain; spectral classification for material labelling | GIS polygon layer with volume (m³) and estimated mass (tonnes) per stockpile, timestamped |
| Period-on-period volume change | Multi-epoch DSM stack differencing; change magnitude and direction per stockpile polygon | Time-series table and chart showing stockpile drawdown or accumulation between tasking dates |
| Retrospective stockpile history from archive stereo | Archive stereo DSM reconstruction at available acquisition dates; gap-filling by interpolation where cloud obscures epochs | Monthly volume reconstruction report covering project duration, suitable for dispute resolution or covenant review |
| Supply-chain delivery event detection | Step-change detection in volume time series; flags acquisitions showing greater than threshold increase between consecutive epochs | Automated alert with before/after DSM thumbnails and volume delta when a significant delivery event is detected |
| Insurance valuation snapshot | Single-epoch tri-stereo DSM; volume-to-mass conversion using published bulk density ranges per material class; uncertainty bounds reported explicitly | PDF valuation report with volume, mass range, density assumptions, DSM accuracy statement and annotated imagery |
| Occlusion and shadow quality flag | Automated shadow mask derived from solar geometry and DSM; flagging of stockpile faces with greater than 20 percent shadowed area as lower-confidence | Per-stockpile confidence score appended to GIS layer attributes |
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.