Petroleum coke stockpile volumetric estimation at refinery and export terminals
Stereo optical and SAR-derived digital surface models let buyers, traders and regulators measure petroleum coke stockpile volumes at refinery yards and export terminals without setting foot on site. Honest limits apply: pile-shape assumptions, moisture and compaction all introduce uncertainty that must be declared.
Sensors
- Pléiades Neo stereo: 30 cm native resolution in panchromatic, in-track stereo pair acquired in a single pass. Generates DSMs with vertical accuracy in the 20–50 cm range over flat industrial surfaces when ground control is available. Revisit of 1–2 days at mid-latitudes.
- WorldView-3 stereo: 31 cm panchromatic, 1.24 m multispectral including SWIR bands. Stereo DSM vertical accuracy similar to Pléiades Neo. SWIR bands (1195–2365 nm range) aid spectral separation of petcoke from coal and ore.
- TanDEM-X: X-band SAR interferometric DSM. Global 12 m posting public product; commercial 2 m CoSSC data available for change-detection DSMs. Cloud-independent. Petcoke's high carbon content and surface roughness produce strong X-band backscatter, typically 3–6 dB above surrounding concrete aprons.
- Sentinel-2 MSI: 10 m visible and NIR, 20 m SWIR, 5-day revisit at equator (2–3 days with both satellites). Too coarse for precise volume integration but useful for pile-extent mapping, change flagging and spectral classification between petcoke, coal and ore.
What a black pile of carbon tells a satellite
Petroleum coke is the residue left after upgrading heavy crude fractions in delayed coking or fluid coking units. It is almost pure carbon, which gives it a very low visible-band reflectance, typically 3–6 % in the red channel, lower than bituminous coal at 5–10 % and far below iron ore or limestone. That spectral contrast is detectable in WorldView-3 SWIR bands and, at coarser scale, in Sentinel-2 Band 11 and Band 12. Published multispectral classification studies have used this signature to separate petcoke from co-located coal and aggregate stockpiles at mixed-commodity terminals with overall accuracies above 85 % when training data are available.
In SAR, the story is backscatter intensity. Petcoke piles are electrically lossy at X-band, but their rough, granular surface texture generates strong volume and surface scattering. TanDEM-X bistatic acquisitions over petcoke yards show backscatter values that are consistently elevated relative to smooth concrete or calm water, making pile boundaries distinguishable even without optical confirmation. This matters at terminals in the Persian Gulf or South-East Asia where haze and cloud cover can interrupt optical tasking for days at a time.
Building the digital surface model and integrating volume
Stereo photogrammetry from Pléiades Neo or WorldView-3 produces a dense point cloud from which a digital surface model is derived at roughly 50 cm post spacing after filtering. Volume is computed by differencing the pile DSM against a bare-earth reference surface, either a pre-stockpile baseline DSM or a modelled ground plane interpolated from surveyed apron points around the pile perimeter. The integration is straightforward in principle; the complication is pile-shape assumption at the edges, where the DSM slope meets the ground plane. An error of 10 cm in the assumed toe elevation across a 200 m pile perimeter propagates into a volume error of several hundred cubic metres.
TanDEM-X interferometric DSMs offer a cloud-independent alternative. The publicly released global TanDEM-X DEM has a 12 m posting and a relative vertical accuracy of approximately 2 m at the 90th percentile, which is too coarse for small piles but adequate for stockpiles exceeding 50 000 tonnes. Commercial bistatic pairs processed to 2 m resolution improve this substantially. A practical workflow pairs a TanDEM-X change-detection DSM for cloud-affected periods with stereo optical DSMs for high-precision snapshots, using the SAR product to bridge gaps rather than as the primary measurement.
The density problem: from cubic metres to tonnes
Volume in cubic metres is not what a commodity trader or port operator needs. They need tonnes. Petcoke bulk density varies considerably: anode-grade (green) petcoke typically runs 700–900 kg/m³ in a freshly deposited pile; fuel-grade petcoke, which is finer and more porous, sits closer to 600–800 kg/m³. Compaction under its own weight increases density with pile height, and rain-saturation can add 5–15 % to apparent density in open storage. None of these parameters are directly observable from orbit.
The honest position is that satellite-derived volume estimates carry a density conversion uncertainty of roughly ±10–20 % unless the terminal operator supplies grade and moisture data. That uncertainty is not a reason to avoid the method; it is a reason to present results as a range rather than a point estimate, and to use consistent density assumptions across time so that change detection remains meaningful even when absolute tonnage is uncertain. A terminal that holds 200 000 m³ of petcoke and grows to 260 000 m³ over six weeks has demonstrably accumulated stock regardless of whether the density is 750 or 850 kg/m³.
Revisit, cloud and the practical monitoring cadence
A single stereo tasking event over a major export terminal costs a commercial tasking slot and delivers one snapshot. For commodity intelligence, what matters is change between snapshots. A monthly cadence is achievable with Pléiades Neo or WorldView-3 at most terminals worldwide, weather permitting. In persistently cloudy regions such as the Gulf of Guinea or parts of South-East Asia, cloud-free stereo acquisition may be possible only four to six times per year from optical sensors alone.
Sentinel-2 at 10–20 m resolution cannot resolve pile geometry for volume integration, but it can detect pile-extent changes of several thousand square metres and flag events worth tasking. A practical architecture uses Sentinel-2 as a continuous watch layer, triggering commercial stereo tasking when the pile footprint changes by more than a threshold. TanDEM-X or Sentinel-1 SAR fills the cloud gaps with lower-precision volume estimates. The result is a time series with varying precision at different epochs, which is more useful than a time series with high precision at only four dates per year.
Accuracy floors and what the method cannot do
Stereo DSM volume estimation over industrial stockpiles has been validated in published literature against ground survey and LiDAR. Relative volume errors of 3–8 % are reported for well-controlled sites with ground control points and known pile geometry. At sites without ground control, errors of 10–15 % are more realistic. These figures assume the pile is fully visible, meaning no conveyor shadow, no adjacent structure occluding the pile edge, and no active loading or unloading during image acquisition.
SAR-based DSMs introduce additional ambiguity at pile edges due to layover and foreshortening effects, particularly on the side of the pile facing the radar look direction. Pile slopes steeper than the incidence angle, common on freshly deposited petcoke cones, can produce layover artefacts that cause the DSM to underestimate height on one flank. Acquisitions from ascending and descending orbits mitigate this but double the data cost. Neither stereo optical nor SAR methods can distinguish petcoke from coal by volume alone; spectral classification from multispectral imagery is a necessary companion step when the terminal handles mixed commodities.
Satellize applies this workflow operationally, combining open Sentinel-2 monitoring with commercial stereo tasking on client licence. The Tonga crop-estimation programme uses a structurally similar DSM-differencing approach for canopy and field-level change, which informs how the team handles ground-plane interpolation in complex terrain.
Typical figures
| Stereo DSM spatial resolution (Pléiades Neo / WorldView-3) | 50 cm post spacing typical after photogrammetric processing |
| Stereo DSM vertical accuracy (with ground control) | 20–50 cm RMSE over flat industrial surfaces |
| TanDEM-X DSM resolution (commercial bistatic) | 2 m posting; global public product at 12 m |
| TanDEM-X relative vertical accuracy (global product) | ~2 m at 90th percentile; better over smooth terrain |
| Sentinel-2 spectral bands used for classification | Band 4 (665 nm), Band 11 (1610 nm), Band 12 (2190 nm) at 10–20 m |
| Minimum pile area for stereo volume integration | ~500 m² practical floor at 50 cm DSM resolution |
| Revisit cadence (commercial stereo tasking) | 1–2 days possible; monthly cadence typical for routine monitoring |
| Volume-to-mass density conversion uncertainty | ±10–20 % without operator-supplied grade and moisture data |
| Archive depth (Pléiades / WorldView tasking archive) | Pléiades from 2012; WorldView series from 2007 |
| Delivery formats | GeoTIFF DSM, vector pile boundary, CSV volume time series, PDF report |
Analytics Satellize can run
| Pile volume snapshot | Stereo photogrammetric DSM differenced against bare-earth reference; volume integrated by trapezoidal or prismatoid rule | GeoTIFF DSM + CSV table of pile ID, area, mean height, volume (m³), estimated mass range (t) |
| Volume change time series | Multi-date DSM stack with consistent ground-plane reference; change flagged at pile level | Time-series chart and GIS layer showing accumulation or drawdown events per pile, with acquisition-date metadata |
| Petcoke vs. coal vs. ore spectral classification | Supervised classification on WorldView-3 SWIR or Sentinel-2 Band 11/12 using published spectral libraries and site-specific training samples | Classified raster and per-class area table; updated on each cloud-free Sentinel-2 pass |
| SAR-based volume estimate (cloud-gap fill) | TanDEM-X bistatic interferometric DSM differencing; pile boundary from co-registered Sentinel-1 backscatter | Volume estimate with explicit uncertainty range flagged as SAR-derived in the time-series feed |
| Sentinel-2 change-flag alert | Pile-footprint extent tracking using NDVI-inverted dark-surface mask; area threshold trigger | Automated alert with thumbnail and area-change figure, used to schedule commercial stereo tasking |
| Density-adjusted mass estimate | Volume estimate combined with operator-supplied or grade-inferred bulk density; Monte Carlo uncertainty propagation over density range | Mass estimate with 80 % confidence interval, reported in tonnes, delivered as PDF and API JSON |
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.