Bulk commodity stockpile volume estimation at export terminals
Combining high-resolution optical footprints with spaceborne elevation data lets analysts estimate coal, iron ore, grain and bauxite stockpile volumes at export terminals, turning pixel counts into inventory and throughput proxies.
Sensors
- TanDEM-X (TerraSAR-X add-on Digital Elevation Measurement): X-band InSAR pair producing digital elevation models at 12 m posting in standard mode, down to approximately 2 m horizontal resolution in high-resolution spotlight mode. Repeat-pass coherence over stable stockpile surfaces is generally good, but loose, freshly deposited material can decorrelate. Tasked acquisitions are required; not a free open archive.
- ICESat-2 (Advanced Topographic Laser Altimeter System, ATLAS): Photon-counting lidar with approximately 17 m along-track beam spacing and sub-decimetre vertical precision on flat surfaces. Revisit at mid-latitudes is roughly 91 days for exact-repeat ground tracks, so coverage of a specific terminal depends on track geometry. Provides authoritative height tie-points for validating or correcting radar-derived DEMs.
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral at nadir. Used for planimetric pile footprint delineation and, via stereo triplet acquisition, for photogrammetric surface models with approximately 0.3 m vertical accuracy under good geometry. Cloud cover and tasking lead time are the practical constraints.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral. Tri-stereo acquisition mode supports photogrammetric DSMs with reported vertical accuracy of around 0.5 m RMSE. Daily revisit capacity over any point. Useful when TanDEM-X tasking windows do not align with operational needs.
- Sentinel-2 (MSI): 10 m multispectral, 5-day revisit at mid-latitudes with two satellites. Insufficient resolution to delineate individual piles at most terminals, but useful for change detection at large facilities and for cloud-screening to identify acquisition windows for higher-resolution tasking.
What a pile of coal is actually hiding
A bulk terminal stockpile is a financial instrument that happens to be made of rock. The gap between what a shipper declares and what is physically present at any moment is the number that traders, lenders, port authorities and commodity desks all want independently. Satellite-derived volume estimation is one of the few methods that does not depend on the terminal operator's own weighbridge or inventory system.
The physics is straightforward. Volume is the integral of height above a reference surface over the pile footprint. Measure the footprint in plan, measure the height field across it, multiply. The difficulty is that neither the footprint nor the height field is trivial to extract from space, and the errors compound.
How elevation gets extracted from orbit
Two sensor families do the heavy lifting. Interferometric SAR, most practically TanDEM-X in tasked spotlight mode, measures phase differences between two radar acquisitions to reconstruct surface height. The German Aerospace Centre (DLR) quotes height accuracy of around 2 m for the global TanDEM-X DEM product, but tasked high-resolution acquisitions over cooperative surfaces can do considerably better. The catch is coherence: freshly deposited, loose material scatters radar energy unpredictably, and a pile that was reshaped between the two passes in a repeat-pass configuration will produce artefacts rather than a clean surface model.
Stereo photogrammetry from very-high-resolution optical satellites is the practical alternative when cloud cover cooperates. WorldView-3 triplet stereo and Pléiades Neo tri-stereo both generate dense point clouds from which a digital surface model is extracted. Vertical accuracy is a function of base-to-height ratio and ground control; over a well-surveyed terminal, 0.3 to 0.5 m RMSE is achievable. ICESat-2 ground tracks, where they intersect the terminal, provide independent lidar height profiles that can validate or calibrate either approach. ICESat-2 is not a mapping instrument, so it cannot generate a full surface model over a terminal on a single pass, but its photon-counting precision makes it an excellent reference.
A third approach, less discussed but worth knowing, is shadow-based height estimation from very-high-resolution optical imagery. If the solar elevation angle at acquisition time is known precisely, the shadow cast by a pile onto a flat apron gives a direct geometric height estimate. This works best for conical piles with clean shadow edges. It fails for irregular or merged piles, and it requires that the shadow falls on open ground rather than on another pile or a structure.
The shape-assumption problem nobody advertises
Every volume calculation rests on an assumption about what lies between the measured points. For a perfect cone, that assumption is trivial. Real stockpiles are not cones. Iron ore piles built by stacker-reclaimers tend to be elongated windrows with relatively predictable cross-sections. Coal piles are often irregular, merged into neighbours, or actively being reshaped during the acquisition window. Grain in covered storage is invisible to optical sensors entirely.
When elevation data is sparse, analysts interpolate. The interpolation error can be a significant fraction of the total volume, particularly for piles with concave surfaces or internal voids. Published studies using airborne lidar as ground truth against satellite-derived volumes report errors in the range of 5 to 15 percent for well-separated, regular piles under good conditions. Irregular or merged piles can exceed 20 percent error. These are not reasons to avoid the method; they are reasons to quote confidence intervals rather than point estimates, and to be explicit about which pile geometries in a given terminal are reliable and which are not.
Turning two snapshots into a throughput signal
A single volume estimate tells you how much is there. Two estimates, separated by a known interval, tell you the net change. Net change is not throughput, but it constrains it. If a terminal receives 200,000 tonnes by rail and the stockpile grows by 150,000 tonnes, approximately 50,000 tonnes departed by vessel in that window. The satellite does not see the rail wagons or the ship's draft directly; it sees the pile, and the pile is the accounting residual.
Revisit frequency therefore matters enormously for operational utility. A 91-day ICESat-2 repeat is useful for calibration but not for weekly inventory tracking. TanDEM-X tasking can, in principle, be scheduled at intervals of days, but cost and scheduling availability are real constraints. Stereo optical acquisitions are weather-dependent; a terminal in a frequently overcast region may yield only a handful of usable stereo pairs per year. For high-cadence monitoring, the honest answer is that the method works best as a periodic audit rather than a continuous feed.
What the method cannot do
Covered storage is invisible. A significant share of grain, and some coal, is stored under roofed structures or in silos. Satellite-derived volume estimation applies only to open stockpiles. Density is assumed, not measured: converting volume to mass requires a bulk density figure, which varies by commodity grade, moisture content and compaction. Wet coal is denser than dry coal. The analyst applies a published or operator-supplied density, and any error in that figure flows directly into the mass estimate.
Temporal ambiguity is a further limit. The satellite captures a moment; if a pile was built up and drawn down between two acquisitions, the net change understates gross throughput. For terminals with very high turnover relative to the revisit interval, the volume-change proxy can be misleading in both directions.
Satellize runs stockpile volume analytics on commercial tasking from WorldView-3 and Pléiades Neo, combined with TanDEM-X or ICESat-2 elevation layers, and delivers change reports as GIS-ready raster and vector products. The methodology is the same class used in the Tonga crop-estimation programme: rigorous uncertainty quantification alongside the headline figure.
Practical accuracy benchmarks from the public record
DLR's published TanDEM-X global DEM specification quotes a relative height accuracy of better than 2 m at 90 percent confidence for non-forest terrain. For stockpile applications using tasked high-resolution spotlight acquisitions, the operational envelope is tighter: independent validation studies in the remote sensing literature report volume errors of roughly 5 to 12 percent against ground survey for well-defined piles. ICESat-2 ATLAS vertical precision over flat, high-reflectance surfaces (which coal and iron ore approximate) is reported by NASA as better than 3 cm for smooth terrain, though real pile surfaces introduce additional scatter.
WorldView-3 stereo DSMs have been validated at sub-metre vertical accuracy in multiple published studies under controlled conditions. In operational use over industrial surfaces with limited ground control, 0.5 to 1 m vertical RMSE is a more conservative and honest expectation. At a typical pile height of 10 to 20 m, that translates to a height uncertainty of 2.5 to 10 percent before any shape-assumption error is added.
Typical figures
| Planimetric resolution (optical) | 0.30 m (WorldView-3 pan), 0.30 m (Pléiades Neo pan) |
| Elevation model posting (TanDEM-X spotlight) | Approximately 2 m horizontal; ~2 m relative height accuracy (DLR specification) |
| Stereo DSM vertical accuracy | 0.3 to 1.0 m RMSE depending on base-to-height ratio and ground control availability |
| ICESat-2 vertical precision | Sub-decimetre on smooth surfaces; 91-day exact-repeat cycle limits spatial coverage per pass |
| Optical revisit (tasked VHR) | 1 to 3 days for WorldView-3 and Pléiades Neo under clear sky; weather-dependent |
| Minimum detectable pile height change | Approximately 0.5 to 2 m depending on sensor and surface coherence; not suitable for shallow change detection |
| Volume estimation error (regular piles, good conditions) | 5 to 15 percent; irregular or merged piles may exceed 20 percent |
| Archive depth | TanDEM-X: tasked acquisitions from 2010; WorldView-3: from 2014; Pléiades Neo: from 2021 |
| Delivery formats | GeoTIFF DSM, Shapefile or GeoPackage footprint polygons, CSV volume table, PDF change report |
| Cloud cover impact | Optical stereo blocked by cloud; TanDEM-X and SAR methods are cloud-transparent |
Analytics Satellize can run
| Pile footprint map | Object-based image analysis and supervised classification on VHR optical imagery to delineate individual pile boundaries | GeoPackage polygon layer with commodity type label and planimetric area per pile, updated per acquisition |
| Surface elevation model | Photogrammetric dense matching (stereo or tri-stereo) or TanDEM-X InSAR processing referenced to a surveyed terminal datum | GeoTIFF DSM at 0.5 to 2 m posting, with uncertainty raster |
| Pile volume estimate | Integration of height above reference plane over footprint polygon, with shape-interpolation uncertainty bounds | CSV table of volume per pile per epoch, with stated confidence interval |
| Volume change and throughput proxy | Differencing of co-registered DSMs between two epochs; net volume change converted to mass using commodity-specific bulk density | Bi-temporal change report (PDF and GIS layer) showing drawdown or accumulation per pile |
| Terminal-level inventory index | Summed volume across all identified piles, normalised to terminal design capacity where publicly documented | Time-series chart and data feed showing aggregate inventory level as percentage of observed maximum |
| Anomaly alert | Statistical threshold on volume change rate relative to historical distribution for the terminal | Email or API alert when single-epoch drawdown or accumulation exceeds a defined threshold |
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.