Coal stockpile volumetric estimation at ports and power stations
Tri-stereo optical imagery and X-band SAR digital surface models can resolve coal stockpile volumes to within roughly one metre vertically, making week-on-week inventory changes visible without a single site visit.
Sensors
- Airbus Pléiades Neo (tri-stereo): 30 cm native resolution in panchromatic; tri-stereo acquisition (three overlapping passes in a single overfly) produces a dense point cloud from which a digital surface model accurate to roughly 0.5–1 m vertically can be derived. Revisit at any given site is typically 1–3 days at mid-latitudes when tasked.
- Maxar WorldView-3 (stereo or tri-stereo): 31 cm panchromatic, 1.24 m multispectral. Stereo pairs yield DSMs with vertical RMSE in the 0.5–1.5 m range over flat industrial yards. Eight SWIR bands also discriminate coal from surrounding materials by reflectance signature.
- ICEYE X-band SAR: Spotlight mode delivers approximately 0.5 m range resolution regardless of cloud or darkness. Repeat-pass interferometry or radargrammetry from ICEYE's constellation (30+ satellites) can produce surface height estimates, though SAR-based DSM accuracy over rough pile surfaces is typically 1–3 m vertical, somewhat coarser than optical tri-stereo.
- Planet SkySat: 50 cm resolution, capable of collect-and-stare video or stereo. Useful for rapid change detection and footprint mapping between higher-accuracy DSM acquisitions; vertical accuracy from SkySat stereo is generally 1–2 m, adequate for detecting large inventory swings.
What a pile's geometry actually tells you
A coal stockpile is, geometrically, a truncated cone or ridge. Its volume is a straightforward integral of height above a known ground plane across the measured footprint. That ground plane is established once from a bare-earth survey or a pre-stockpile DSM; subsequent acquisitions measure the current surface. Subtract one from the other, integrate, apply a bulk density figure (bituminous coal typically runs 800–900 kg/m³ in an unconsolidated yard pile), and you have a tonnage estimate.
The physics is not subtle. The uncertainty sits in three places: the vertical accuracy of the DSM, the assumed bulk density, and the fraction of the pile base obscured by conveyor infrastructure. Tri-stereo photogrammetry from Pléiades Neo handles the first problem well. The density assumption is the honest weak point: operators who share their material specification can tighten it considerably.
Why tri-stereo outperforms a simple stereo pair
A conventional stereo pair gives two views of the same surface at slightly different angles. Matching errors in low-texture regions, such as the dark, uniform face of a coal pile, degrade the point cloud. Tri-stereo adds a third image from a third angle in the same overfly, giving the photogrammetric engine a third constraint that resolves ambiguities and fills shadow zones on the lee side of tall piles.
Airbus publishes vertical accuracy figures of around 0.5 m RMSE for Pléiades Neo tri-stereo over well-controlled industrial sites. At a large export terminal where individual piles may hold 50,000 to 200,000 tonnes, a one-metre vertical error across a pile footprint of 5,000 m² translates to a volume uncertainty of roughly 5,000 m³, or about 4,000–4,500 tonnes. That is a few percent of pile mass, which is commercially meaningful but not disqualifying for weekly trend analysis.
WorldView-3 performs comparably. The choice between them is usually a scheduling question: which constellation can task the site on the required day.
SAR as a cloud-proof alternative, with honest caveats
Optical tri-stereo is elegant but helpless under cloud. Ports in South-East Asia, West Africa, and the Pacific Northwest can run cloud-free windows of only a few days per month. ICEYE X-band SAR fills that gap: it images through cloud and at night, and its spotlight mode resolves individual pile ridges.
The limitation is real, though. Radar backscatter from a coal pile depends on surface roughness and moisture as much as geometry. Radargrammetry from two SAR acquisitions at different incidence angles can produce a DSM, but vertical accuracy over a rough, heterogeneous pile surface is typically 1–3 m rather than the sub-metre figures achievable optically. For a small pile this matters; for a large terminal stockpile it is often acceptable. A sensible operational design uses optical DSMs as the primary volumetric measurement and SAR as a between-acquisition change-detection layer, flagging significant pile-shape changes that warrant a new optical task.
The footprint problem and how to handle it
Volume calculation requires an accurate pile footprint as well as a height model. Automated extraction from 30 cm imagery works well for isolated piles but struggles where piles abut each other or share a base. The standard approach is to delineate pile boundaries in the first high-resolution acquisition by hand or with supervised segmentation, then propagate those boundaries forward using change in surface height as the guide.
Conveyor galleries and stacker-reclaimer machines that sit on or against a pile introduce artefacts. A stacker arm at 15 m elevation will appear in the DSM as a false height spike. Post-processing filters that compare DSM values against expected pile slope angles (typically 35–40 degrees for coal at the angle of repose) catch most of these. Residual artefacts are flagged for analyst review rather than silently discarded.
The same workflow applies without modification to iron-ore, bauxite, and phosphate stockpiles. Bulk density changes, pile colour changes, and the analyst updates the material specification. The geometry is identical.
Frequency, latency, and what the market actually needs
Commodity traders and power-station fuel managers rarely need real-time data. What they need is a reliable weekly or fortnightly snapshot with a latency of 24–48 hours from acquisition to delivered volume figure. That cadence is achievable with commercial tasking on Pléiades Neo or WorldView-3, subject to cloud and scheduling.
For power stations, the critical question is days-of-supply remaining, which requires dividing current inventory by daily consumption rate. The satellite provides the inventory; the consumption rate comes from plant operating data or from a parallel thermal-infrared observation of stack activity (covered on the sibling page for refinery unit utilisation signals). The two signals together give a more complete picture than either alone.
Satellize structures analytics engagements around a defined monitoring cadence rather than ad-hoc tasking, which keeps scheduling priority consistent and avoids the gaps that occur when sites are only tasked reactively. The same operational discipline that underpins the Tonga crop-estimation programme applies here.
Accuracy limits every buyer should understand
No satellite-derived volume figure is as accurate as a laser survey or a belt-scale integration. The honest comparison is with the alternatives buyers actually use: visual estimation from CCTV or drone overflights (highly variable), periodic manual surveys (infrequent and expensive at remote terminals), and shipper declarations (subject to obvious incentive problems).
Satellite photogrammetry at 30 cm resolution, properly processed, achieves vertical RMSE of 0.5–1.5 m depending on site conditions and pile texture. Over a large pile this yields a volume uncertainty of roughly 2–5 percent. Over a small pile of 5,000 tonnes the percentage error is larger and the method is less useful. Cloud cover at tropical ports can interrupt the optical acquisition cadence for weeks; SAR backup mitigates but does not eliminate this. Ground control points, ideally surveyed targets painted on the yard surface, reduce systematic DSM bias and are worth the modest one-time cost.
Typical figures
| Best available spatial resolution (optical) | 30 cm (Pléiades Neo panchromatic) |
| Vertical DSM accuracy, optical tri-stereo | 0.5–1.5 m RMSE over flat industrial yards |
| Vertical DSM accuracy, X-band SAR radargrammetry | 1–3 m typical over rough pile surfaces |
| Revisit (tasked commercial optical) | 1–3 days at mid-latitudes; longer at high latitudes |
| All-weather / night-time capability | X-band SAR (ICEYE spotlight); optical blocked by cloud |
| Minimum pile size for reliable volume estimate | Approximately 10,000 m² footprint; smaller piles carry higher percentage error |
| Typical volume uncertainty (large pile, good conditions) | 2–5 percent of total pile volume |
| Archive depth (Pléiades / SPOT heritage) | Pléiades archive from 2012; SPOT 5 heritage extends to early 2000s |
| Delivery formats | GeoTIFF DSM, point cloud (LAS/LAZ), polygon volume report (CSV/PDF), GIS-ready pile-boundary shapefile |
| Spectral bands available (WorldView-3) | Panchromatic, 8-band VNIR, 8-band SWIR (1195–2365 nm) |
Analytics Satellize can run
| Weekly stockpile volume and tonnage estimate | Photogrammetric DSM differencing from tri-stereo optical imagery; volume integration over delineated pile footprint; bulk-density conversion | PDF report and CSV table with per-pile volume, estimated tonnage, and week-on-week change; GeoTIFF DSM included |
| Pile footprint and boundary map | Supervised segmentation on 30 cm panchromatic imagery, refined by slope-angle filtering of DSM | Shapefile or GeoJSON polygon layer, updated each acquisition cycle |
| Between-acquisition change alert | SAR backscatter change detection (ICEYE spotlight repeat pass); significant pile-shape change triggers optical re-task | Automated alert (email or API) with flagged site, estimated magnitude of change, and recommended next-task date |
| Historical inventory time series | Archive DSM reconstruction from Pléiades / SPOT / WorldView archive stereo pairs; gap-filled with SAR-derived relative change | Time-series chart (PNG/interactive HTML) and underlying data table (CSV) covering available archive period |
| Material classification layer | WorldView-3 SWIR band-ratio analysis to discriminate coal from ore, aggregate, or bare ground by spectral reflectance signature | Classified raster and per-class area statistics; useful for mixed-commodity yards |
| Days-of-supply estimate (power stations) | Satellite-derived inventory volume combined with client-supplied or independently estimated daily consumption rate | Weekly briefing note with current inventory, consumption assumption, implied days-of-supply, and confidence range |
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.