Construction material stockpile volume tracking on active sites
Photogrammetric digital surface models derived from satellite stereo pairs let lenders and project monitors measure stockpile volumes independently, catching discrepancies between materials on site and certified drawdowns before funds are released.
Sensors
- Airbus Pléiades (tri-stereo): 0.5 m panchromatic resolution; tri-stereo acquisition in a single pass generates DSMs with vertical accuracy typically 0.3–0.5 m RMSE under good ground-control conditions, sufficient to resolve stockpiles above roughly 1 m height. Revisit is 1–2 days at mid-latitudes via the Pléiades constellation.
- Planet SkySat (video-derived DSM): 0.5 m native resolution; video mode produces overlapping frames from a single overpass that can be processed into dense point clouds via structure-from-motion. Vertical accuracy is somewhat lower than dedicated stereo systems, typically 0.5–1.0 m, but revisit can be daily and tasking latency is short.
- Maxar WorldView stereo (WorldView-2/3): WorldView-3 delivers 0.31 m panchromatic imagery; in-track stereo pairs yield DSMs with sub-metre vertical accuracy. The 8-band SWIR capability on WorldView-3 can additionally help distinguish material types by surface reflectance, though that is secondary to the geometric product.
- Capella Space SAR (X-band): Spotlight mode achieves 0.5 m resolution; SAR is cloud-independent, which matters on sites where optical revisit is broken by overcast. Radargrammetric DSMs from SAR stereo pairs are less precise than optical photogrammetry (vertical errors of 1–2 m are common), but the all-weather availability makes it a useful gap-filler between optical acquisitions.
What a pile of sand can tell a lender
Construction finance relies on drawdown certificates: a contractor declares that materials worth a stated sum are on site, and funds are released accordingly. The certificate is a document. The stockpile is a physical object with a measurable volume. Satellite stereo imagery closes the gap between the two.
Volume is computed by differencing a current digital surface model against a pre-construction terrain baseline. The baseline is typically acquired before ground is broken, or reconstructed from archival tasking. Every subsequent stereo acquisition produces a new DSM; the signed volume difference at each stockpile location is the net accumulation or drawdown since the last epoch. A lender receives a figure in cubic metres, not a photograph.
How stereo photogrammetry turns two images into a surface model
A stereo pair is two images of the same scene taken from slightly different angles, either in a single pass (in-track stereo) or across passes (cross-track). Dense image matching algorithms, most commonly semi-global matching (SGM) or its variants, find corresponding pixels across the pair and triangulate their three-dimensional positions. The output is a point cloud, which is then gridded into a raster DSM at a post-spacing typically equal to two to four times the image ground sampling distance.
For Pléiades tri-stereo at 0.5 m GSD, the resulting DSM is commonly posted at 0.5–1.0 m. A conical aggregate stockpile 20 m in diameter and 5 m tall contains roughly 1,000 cubic metres; at that scale, a vertical error of 0.5 m translates to a volume uncertainty of perhaps 5–8 per cent, which is acceptable for financial verification purposes. Smaller piles, say under 3 m in any dimension, approach the resolution floor and should be flagged with wider uncertainty bounds rather than a point estimate.
Ground control points, either surveyed targets placed on site or identifiable stable features used as tie points, are the primary lever for improving absolute vertical accuracy. Without any ground control, satellite DSMs can carry systematic offsets of 1–3 m. With even a handful of well-distributed GCPs, that offset collapses to the sub-half-metre range.
Cloud, shadow and the honest limits of optical stereo
Optical stereo fails under cloud cover. On a site in a maritime climate, a programme relying solely on Pléiades may find that two or three successive tasking attempts are clouded out, creating gaps of several weeks in the monitoring record. This is not a solvable problem with better algorithms; it is physics. The mitigation is to maintain a parallel SAR tasking thread, accepting the lower vertical precision of radargrammetry in exchange for cloud-independence.
Shadow is a subtler issue. Stockpiles near tall structures or acquired at low solar elevation cast shadows that confuse the image-matching step, producing voids or artefacts in the DSM. Processing protocols should mask shadow regions and report them as no-data rather than interpolating across them. A stockpile that is 40 per cent shadowed cannot be reliably volumetised from that epoch.
Material type also matters. Wet sand and dry aggregate have different surface textures and reflectances. Neither causes a fundamental problem for photogrammetry, but specular reflection off wet surfaces can degrade matching quality. SkySat video mode, because it samples many viewing angles rapidly, is somewhat more tolerant of surface glint than a two-image stereo pair.
Building the monitoring workflow
A practical monitoring programme for a large infrastructure site typically runs on a two-to-four-week cadence, aligned with drawdown certificate submission dates. Each epoch requires a tasking order, an image quality check, DSM generation, co-registration to the baseline terrain, stockpile delineation (either from a pre-defined polygon layer or automated change detection), volume computation and a structured report.
Stockpile delineation deserves attention. Manually drawn polygons, agreed with the site team at programme outset, are the most defensible approach for financial reporting: both parties know exactly which heap is being measured. Automated detection using DSM differencing thresholds works well for large, isolated piles but can merge adjacent stockpiles or clip irregular ones. On a complex site with many material types in close proximity, a hybrid approach, automated detection reviewed by an analyst, is more reliable than either extreme.
Archive depth matters for baseline construction. Pléiades has been operational since 2011 and 2012 for the two satellites; WorldView-2 since 2009. For sites where pre-construction tasking was not commissioned, it is often possible to reconstruct a pre-disturbance terrain from archive imagery, though archive coverage of any specific site is not guaranteed and should be confirmed before a programme is scoped.
What SAR adds when clouds close in
Capella Space operates X-band SAR satellites capable of 0.5 m spotlight imagery. X-band backscatter is sensitive to surface roughness and dielectric properties, which means coarse aggregate, smooth sand and waterlogged spoil return different signal intensities. That is useful for qualitative material discrimination, though not a substitute for laboratory analysis.
Radargrammetry, using two SAR images from different incidence angles, produces DSMs with vertical accuracy typically in the 1–2 m range for well-separated baselines. That is sufficient to detect the appearance or disappearance of a large stockpile between epochs, but not to produce the sub-metre volume precision that optical stereo achieves under clear skies. The honest workflow treats SAR as continuity insurance: it keeps the monitoring record intact through cloudy periods and flags significant changes that warrant a follow-up optical acquisition when conditions clear.
Delivering the number a project monitor can sign off
The output that matters to a quantity surveyor or project finance team is not a point cloud. It is a volume figure with an uncertainty range, a map showing which polygons were measured and which were flagged as unreliable, and a change chart showing accumulation and drawdown across the programme timeline. That package needs to be reproducible: the same inputs should yield the same outputs, and the methodology should be documented well enough that an independent reviewer could audit it.
Satellize structures stockpile monitoring reports to match the drawdown certificate schedule, with each epoch report citing the DSM acquisition date, cloud cover percentage over the site, ground control configuration, per-stockpile volume estimate and associated uncertainty, and a comparison against the certified quantity. The Tonga crop-estimation programme established the same principle in a different domain: the analytic output is only as useful as its traceability. A volume figure without a documented uncertainty is not a verification; it is a guess dressed up as one.
For sites where the monitoring programme is agreed at financial close, Satellize can also provide a tasking schedule aligned to the drawdown calendar, so that imagery is acquired and processed before each certificate submission rather than retrospectively.
Typical figures
| Typical DSM post-spacing | 0.5–1.0 m (Pléiades, SkySat, WorldView stereo); 1–2 m (SAR radargrammetry) |
| Vertical accuracy (with GCPs) | 0.3–0.5 m RMSE for optical stereo; 1–2 m for SAR radargrammetry |
| Minimum reliably volumetised stockpile | Approximately 3 m height and 10 m diameter; smaller piles carry uncertainty exceeding 20 per cent |
| Revisit cadence | 1–2 days (Pléiades constellation, subject to cloud and tasking priority); daily potential (SkySat); SAR revisit 1–3 days (Capella) |
| Cloud impact | Optical stereo unusable under cloud cover; SAR unaffected |
| Archive depth | Pléiades from 2011–12; WorldView-2 from 2009; archive coverage of any specific site not guaranteed |
| Delivery formats | GeoTIFF DSM, LAS point cloud, GeoPackage polygon layer with volume attributes, PDF epoch report, CSV change timeseries |
| Typical programme cadence | Fortnightly to monthly, aligned to drawdown certificate schedule |
| Spectral bands (WorldView-3 supplementary) | 8-band VNIR plus 8-band SWIR for surface material discrimination |
Analytics Satellize can run
| Per-stockpile volume estimate with uncertainty bounds | Semi-global matching DSM differencing against pre-construction terrain baseline; uncertainty propagated from GCP residuals and shadow mask extent | Epoch report (PDF) with tabulated volumes, uncertainty ranges and polygon map |
| Drawdown certificate verification flag | Comparison of computed volume change against certified drawdown quantity; discrepancy expressed as percentage and absolute cubic metres | Structured data table (CSV/Excel) appended to epoch report, flagging material variances above agreed threshold |
| Accumulation and drawdown timeline chart | Multi-epoch volume timeseries aggregated by material type polygon; trend fitted to identify anomalous drawdown events | Interactive chart (HTML) or static figure (PDF) covering full programme period |
| SAR continuity DSM for clouded epochs | Radargrammetric processing of Capella X-band stereo pairs; co-registered to optical DSM coordinate frame | GeoTIFF DSM with accompanying quality note on reduced vertical precision |
| Material-type discrimination layer (WorldView-3 sites) | Spectral classification of SWIR bands to separate aggregate, sand, spoil and concrete by surface reflectance signature | GeoPackage polygon layer with material-type attributes, delivered alongside volume estimates |
| Baseline terrain model | Pre-construction stereo acquisition or archive DSM reconstruction; used as the fixed reference for all subsequent differencing | GeoTIFF DTM archived and versioned for programme duration |
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.