Coal stockpile volume estimation from optical stereo imagery
Tri-stereo and along-track stereo optical imagery can reconstruct coal stockpile volumes to within a few percent under clear skies, giving commodity traders, insurers and lenders an independent tonnage check without setting foot on site.
Sensors
- Pleiades Neo (Airbus): 30 cm native panchromatic resolution; tri-stereo mode acquires fore, nadir and aft views in a single pass, enabling dense point-cloud generation with published vertical accuracy of 0.3–0.5 m RMSE over flat terrain. Revisit of 1–2 days at mid-latitudes.
- WorldView-3 (Maxar): 31 cm panchromatic; along-track stereo pairs collected in a single overpass. Published CE90 horizontal accuracy of roughly 3.5 m without ground control, tightening to sub-metre with GCPs. Useful stereo base-to-height ratio of 0.4–0.6 achievable.
- SPOT 6/7 (Airbus): 1.5 m panchromatic resolution; along-track stereo pairs available. Lower resolution than Pleiades Neo but a large archive stretching back to 2012–2014 supports historical volume reconstruction. Vertical accuracy typically 1–2 m RMSE.
- Pleiades 1A/1B (Airbus archive): 50 cm panchromatic; stereo and tri-stereo archive from 2011 onwards. Provides a decade-long baseline for stockpile change analysis at terminals with long operating histories.
What a pile of coal actually tells a satellite
A coal stockpile is, geometrically, a simple object: a heap of dark granular material sitting on a flat concrete or compacted-earth pad. Its volume is the integral of height above that pad surface. The challenge is that no single nadir image records height. Two images taken from slightly different angles do, because the same point on the pile crest appears at a measurable horizontal offset between the two frames. That offset, called parallax, scales directly with height given known camera geometry. This is photogrammetric stereo, a principle unchanged since the 1930s, now applied from 600 km altitude with sub-metre results.
Modern very-high-resolution satellites such as Pleiades Neo collect a fore image, a nadir image and an aft image in a single uninterrupted pass, typically within 60–90 seconds. Because the acquisition is simultaneous in practice, shadows, lighting and atmospheric conditions are nearly identical across all three frames. That consistency is important: shadow edges on a coal pile can mimic topographic relief if the illumination geometry shifts between acquisitions, which is a real problem for stereo pairs collected on separate orbits.
From pixel parallax to tonnes: the processing chain
The photogrammetric pipeline converts the stereo imagery into a dense point cloud, then interpolates that cloud into a digital surface model (DSM) at roughly 0.5–1 m post spacing for Pleiades Neo data. A baseline DSM of the empty pad, derived from an earlier acquisition or from a survey, is subtracted to isolate the stockpile volume above grade. The resulting volume figure in cubic metres is then multiplied by a bulk density to convert to tonnes.
That density step is where honest uncertainty lives. Thermal coal bulk density varies by rank, moisture content and compaction method, typically falling between 750 and 900 kg per cubic metre for run-of-mine coal and rising toward 950 kg per cubic metre for compacted export-grade product. Without direct sampling, an analyst must assume a density, which introduces a tonnage uncertainty of roughly 5–15% depending on coal type and stockpile management practice. Reporting the volume in cubic metres alongside the tonnage estimate, with an explicit density assumption, is the only intellectually honest approach.
Irregular pile shapes add a secondary complication. Reclaimer equipment creates conical or elongated ridges with steep flanks. Steep flanks cast shadows that can occlude the DSM in the downward-sun direction, creating data voids that must be interpolated. Tri-stereo geometry helps because the fore and aft views illuminate different shadow zones, but it does not eliminate the problem entirely on piles with slopes exceeding roughly 35–40 degrees.
Cloud cover is not a nuisance. It is a hard stop.
Optical stereo is unambiguously blocked by cloud. A single overcast day over a port terminal means no acquisition, no DSM, no volume figure. This is not a minor caveat. Major coal export terminals in Queensland, Indonesia, South Africa and Colombia all experience significant cloud frequency, particularly during wet seasons. At some locations, cloud-free acquisition windows narrow to a handful of days per month.
The practical mitigation is task-and-wait scheduling: a standing tasking order on the target facility, with the operator collecting whenever cloud probability falls below a threshold, typically below 10–20% forecast cloud cover. Turnaround from clear-sky acquisition to delivered volume report is typically 24–72 hours for commercial providers. For time-sensitive applications such as vessel loading verification or month-end inventory reporting, that latency needs to be built into the workflow. SAR-based stockpile estimation, covered in the sibling page on bulk-storage inventory, is the complementary method for cloud-persistent environments, though SAR carries its own shape-reconstruction limitations.
Who actually uses this, and what they need from it
The primary buyers of coal stockpile volume data fall into three groups. Commodity traders and hedge funds want an independent check on reported inventory at major export terminals, particularly when spot prices are moving on supply narratives that cannot be verified from public data. Insurers need to verify declared values for stock-throughput and property policies covering coal yards, where a single large terminal may carry insured values in the hundreds of millions of dollars. Project-finance lenders and bond holders want periodic confirmation that coal assets pledged as collateral actually exist at the declared volume.
Each group has a different tolerance for uncertainty. A trader may accept a 10% volume uncertainty if the direction of change is clear. An insurer writing a policy on a declared 500,000-tonne stockpile needs a tighter figure and will want the density assumption documented. A lender may require a formal survey-grade report with stated confidence intervals rather than an analytic estimate. The satellite method sits between desktop estimation and physical survey: more rigorous than extrapolating from ship manifests, less precise than a licensed surveyor with ground-based LiDAR.
Satellize runs stereo-tasking workflows for clients who need periodic volume snapshots at specific facilities, with outputs delivered as GIS layers and structured reports. The methodology is consistent with the approach used in the Tonga crop-estimation programme, where volume-proxy metrics are derived from high-resolution imagery and reported with explicit uncertainty bounds.
Archive depth and what history reveals
The Pleiades 1A/1B archive runs from 2011 and SPOT 6/7 from 2012–2014, which means historical stereo reconstruction is possible at many major terminals going back roughly a decade. This archive depth is commercially significant. An analyst can reconstruct quarterly stockpile volumes at a facility over several years, cross-reference them with reported export tonnages and AIS-derived vessel call data, and identify systematic discrepancies. Persistent over-declaration of inventory relative to satellite-derived volumes, for instance, is a pattern that has appeared in commodity-finance fraud cases.
Archive tasking was not always stereo-optimised, so historical coverage is patchy. Not every date has a usable stereo pair over a given target. A realistic expectation for a major terminal in a moderately cloudy environment is 4–8 usable stereo acquisitions per year from archive, with the ability to increase that cadence through forward tasking. That is enough for quarterly or bi-monthly inventory tracking but not for daily monitoring.
Typical figures
| Spatial resolution (panchromatic) | 0.30 m (Pleiades Neo), 0.31 m (WorldView-3), 1.5 m (SPOT 6/7) |
| DSM post spacing (typical) | 0.5–1.0 m for Pleiades Neo tri-stereo; 1–2 m for SPOT 6/7 stereo |
| Vertical accuracy (DSM) | 0.3–0.5 m RMSE (Pleiades Neo, with GCPs); 1–2 m RMSE (SPOT 6/7) |
| Revisit (tasked) | 1–2 days (Pleiades Neo); 1–3 days (WorldView-3); 2–3 days (SPOT 6/7) |
| Acquisition latency to delivery | 24–72 hours from clear-sky acquisition to volume report |
| Minimum detectable stockpile | Approximately 1,000–2,000 m³ volume above pad (geometry-dependent) |
| Tonnage uncertainty (density assumption) | ±5–15% depending on coal rank, moisture and compaction |
| Cloud sensitivity | Total blockage; optical stereo requires cloud-free acquisition |
| Archive depth | Pleiades 1A/1B from 2011; SPOT 6/7 from 2012–2014; WorldView from 2014 |
| Delivery formats | GeoTIFF DSM, shapefile/GeoJSON stockpile polygons, PDF/structured report with volume and tonnage table |
Analytics Satellize can run
| Point-in-time stockpile volume and tonnage estimate | Photogrammetric DSM generation from tri-stereo or along-track stereo imagery; baseline subtraction; volume integration | Structured PDF report with volume (m³), tonnage range (with stated density assumption), and annotated orthoimage |
| Change detection: volume delta between two acquisition dates | DSM differencing between two co-registered surface models; change map generation | GeoTIFF difference raster and tabular summary of volume added or removed per stockpile polygon |
| Historical inventory reconstruction | Archive stereo retrieval and batch DSM processing over user-defined date range | Time-series spreadsheet of quarterly or monthly volume estimates with confidence intervals; GIS layer stack |
| Stockpile polygon delineation | Spectral classification of coal (high absorption in VNIR, very low reflectance) combined with DSM-derived boundary | GeoJSON polygon layer of individual stockpile footprints with area and centroid coordinates |
| Declared-versus-observed inventory comparison | Satellite-derived volume cross-referenced against client-supplied manifest or publicly reported inventory data | Discrepancy report flagging facilities where satellite volume falls outside a stated tolerance of declared figures |
| Tasking alert on acquisition success | Cloud-probability monitoring and automatic tasking trigger; quality-check on delivered stereo pair before processing | Email or API notification on successful acquisition with preliminary volume figure within 24 hours |
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.