LNG and petroleum storage tank fill-level estimation
Floating-roof storage tanks reveal their fill level through the shadow cast by the roof rim in sub-metre optical imagery and through roof position in SAR data. The method is well documented, commercially deployable, and honest about its limits.
Sensors
- Maxar WorldView-3: 0.31 m panchromatic resolution; the sharpest commercially available optical imagery for shadow-width measurement. Revisit roughly 1 day at mid-latitudes depending on tasking priority. Cloud is the binding constraint.
- Airbus Pléiades Neo: 0.30 m panchromatic, 1.2 m multispectral. Stereo and tri-stereo modes allow independent height estimation of the roof rim as a cross-check on shadow geometry. Revisit 1 day with the two-satellite constellation.
- ICEYE SAR (X-band): Spotlight mode delivers roughly 0.25 m resolution; stripmap 3 m. SAR penetrates cloud and works at night, detecting the roof position as a bright double-bounce or specular return. Revisit varies by contract but can reach sub-daily over a priority site.
- Capella Space SAR (X-band): Spotlight products at approximately 0.35 m resolution. Useful for detecting gross roof position changes and for monitoring sites where persistent cloud makes optical collection unreliable. Sub-daily revisit available under tasking agreements.
- ESA Sentinel-1 (C-band SAR): Free, 6-day repeat at 10 m IW mode resolution. Insufficient for shadow geometry on individual tanks, but useful for change detection across large terminal clusters and for calibrating trends over time.
What a floating roof gives away
An external floating roof sits directly on the liquid surface and rises or falls with it. The rim of that roof casts a shadow on the inner tank wall whose width is a direct function of how far the roof sits below the tank top. Measure the shadow width in a high-resolution image, know the tank diameter and the solar elevation angle at acquisition time, and you can calculate the liquid depth to within roughly 5 to 15 per cent of tank capacity under good conditions. The geometry is straightforward trigonometry, not inference.
The method was formalised in the remote-sensing literature and is now standard practice among commodity traders and energy analysts. It requires sub-metre imagery because a typical tank rim is only 0.5 to 1.5 metres wide, and the shadow it casts at moderate solar angles may span just 1 to 4 metres. At Landsat or Sentinel-2 resolution, the shadow is simply not resolvable. WorldView-3 and Pléiades Neo sit well inside the required threshold.
SAR as the all-weather alternative
Optical shadow geometry fails under cloud, at night, and when solar elevation is too low or too high to produce a usable shadow. SAR sidesteps all three constraints. In X-band spotlight imagery, the floating roof appears as a bright disc whose position within the tank cylinder can be measured. The roof-to-tank-wall gap produces a characteristic radar return pattern that shifts as the fill level changes.
The honest caveat is that SAR-based fill estimation is less geometrically clean than the shadow method. Roof tilt, surface roughness, and the look-angle geometry all introduce noise. Published studies report uncertainties in the range of 10 to 20 per cent of tank capacity for SAR, compared with 5 to 15 per cent for optical under good conditions. The practical answer is to use both: optical when skies permit, SAR to fill the gaps and to confirm large changes.
What the method cannot do
Fixed-roof tanks are opaque to both approaches. Cone-roof, dome-roof, and underground storage facilities give no external geometric signal of their contents. Internal floating-roof tanks, which have an outer fixed roof over an inner floating pan, are similarly unreadable from orbit.
Tank diameter matters. The shadow method becomes unreliable on tanks smaller than roughly 30 metres in diameter, because the shadow width at partial fill approaches the resolution limit even of WorldView-3. Very large tanks, 80 metres or more in diameter, are the easiest targets. LNG tanks at export terminals are frequently in this range. Solar elevation angle is a further constraint: angles below about 20 degrees produce shadows that spill beyond the tank wall and become unmeasurable, while angles above 70 degrees compress the shadow to near-zero width. Acquisition scheduling must account for this.
From pixel to barrel: the processing chain
The pipeline starts with precise orthorectification, which requires a good digital elevation model of the terminal to remove terrain-induced parallax. The tank rim is then extracted by circular Hough transform or equivalent edge-detection method. Shadow width is measured along multiple azimuths and averaged to reduce noise from non-uniform illumination or partial obstruction by pipework. Solar geometry at the exact acquisition time is applied to convert shadow width to roof depression depth. Tank capacity curves, derived from tank diameter and a standard cylindrical assumption or from operator-published specifications, convert depth to volume.
Uncertainty compounds at each step. Orthorectification error, edge-detection noise, solar ephemeris precision, and tank geometry assumptions each contribute. A well-run pipeline on a large, unobstructed tank in full sun can achieve fill estimates within 5 per cent of capacity. A partially shaded tank with pipework crossing the shadow zone may be accurate only to 20 per cent. Reporting confidence intervals alongside point estimates is not optional; it is what separates a useful product from a misleading one.
Practical applications and the intelligence they support
The primary users are commodity traders tracking crude and refined-product inventories at key terminals, energy ministries monitoring strategic reserve levels, and port operators managing throughput capacity. A time-series of tank fill estimates across a terminal, collected weekly or after each tasking opportunity, produces an inventory trend that is independent of operator disclosures or customs data.
Secondary applications include supply-chain risk assessment: a terminal cluster showing consistently low fill levels across multiple tanks may signal a supply disruption before it appears in official statistics. Conversely, tanks at capacity constrain tanker discharge scheduling. Neither application requires perfect accuracy; directional signals at 10 to 15 per cent precision are commercially actionable.
Satellize applies this method as part of its broader satellite analytics capability, combining open-constellation monitoring with commercial tasking on client licence. Tasking schedules are designed around solar elevation windows and cloud climatology for each terminal location, which varies considerably between, say, the Arabian Gulf and the North Sea.
Calibration, validation and the limits of the public record
Published validation studies have compared shadow-derived estimates against reported inventory figures for terminals in the United States, Europe, and the Middle East, generally finding root-mean-square errors of 5 to 12 per cent of tank capacity for large tanks under good imaging conditions. No published study has achieved better than about 3 per cent, and that figure required ground-truth data that is rarely available commercially.
SAR validation is thinner in the public literature. Most published work uses simulated or controlled scenarios rather than operational terminal data. Treat SAR-derived fill estimates as directional indicators rather than precise measurements until a site-specific calibration dataset can be built. The good news is that with consistent tasking, a calibration history accumulates naturally over time, and estimates improve as the tank's radar signature at known fill levels is characterised.
Typical figures
| Optical spatial resolution (operational floor) | 0.30 to 0.31 m panchromatic (Pléiades Neo, WorldView-3); sub-metre is mandatory for the shadow method |
| SAR spatial resolution (spotlight mode) | 0.25 to 0.35 m (ICEYE, Capella); 10 m Sentinel-1 IW for change detection only |
| Revisit (optical, commercial tasking) | 1 day typical at mid-latitudes; cloud-limited; effective clear-sky revisit varies by climate zone |
| Revisit (SAR, commercial tasking) | Sub-daily to 3 days depending on constellation and contract; cloud-independent |
| Solar elevation window for shadow method | 20 to 70 degrees; acquisitions outside this range are unusable for fill estimation |
| Minimum tank diameter for reliable estimation | Approximately 30 m; larger tanks (60 m+) give substantially lower uncertainty |
| Typical fill-level uncertainty (optical, large tank, good conditions) | 5 to 15 per cent of tank capacity; SAR typically 10 to 20 per cent |
| Archive depth | WorldView: commercial archive from 2014; Sentinel-1: 2014 to present (free); Pléiades: 2011 to present |
| Delivery formats | GeoJSON fill-level time series, GeoTIFF annotated imagery, CSV inventory report, dashboard feed |
| Tank types amenable to method | External floating roof only; fixed-roof and internal floating-roof tanks are not measurable by this technique |
Analytics Satellize can run
| Single-acquisition tank fill estimate | Shadow-width trigonometry on sub-metre optical imagery; circular Hough transform for rim detection | PDF or GeoJSON report with fill level, confidence interval, and annotated image chip per tank |
| Weekly inventory time series | Repeated optical and SAR tasking fused into a continuous fill-level record; gap-filled by SAR when cloud blocks optical | CSV or API feed of dated fill estimates with uncertainty bands, per tank and per terminal aggregate |
| Terminal-level inventory index | Summed fill estimates across all external floating-roof tanks at a terminal, normalised to total nameplate capacity | Weekly dashboard tile showing per cent utilisation and week-on-week change |
| Inventory anomaly alert | Statistical process control on rolling fill-level time series; alert triggered when change exceeds defined threshold | Email or webhook alert with annotated before/after image chips and estimated volume change in barrels or cubic metres |
| Multi-terminal comparative inventory report | Parallel shadow-geometry processing across a defined set of terminals; results ranked by fill level and rate of change | Weekly PDF or spreadsheet comparing inventory levels across a user-defined terminal list |
| SAR-only night or cloud-persistent monitoring | Roof-position detection in X-band spotlight SAR; calibrated against optical baseline where available | GeoJSON fill-level estimates flagged as SAR-derived with wider confidence intervals, integrated into the same time-series feed |
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.