Crude oil tank farm aggregate inventory change from multi-tank shadow analysis
Aggregating shadow-derived fill levels across a tank farm converts noisy single-tank geometry into a site-level inventory estimate. This page explains the statistics, the SAR fallback when cloud blocks optical sensors, and the physical ambiguity that caps accuracy.
Sensors
- Airbus Pléiades Neo: 30 cm panchromatic resolution; shadow widths on tanks as small as 10 m diameter are measurable. Tasking latency typically 24–48 hours. Clear-sky dependent.
- Planet SkySat: 50 cm resolution with same-day tasking possible; revisit constrained by constellation geometry to roughly 2–4 times daily at mid-latitudes. Adequate for tanks above roughly 20 m diameter.
- Capella Space SAR-X: X-band SAR; spotlight mode delivers approximately 0.5 m resolution. Operates through cloud and at night. Roof-deck backscatter and shadow geometry are both usable for fill-level inference.
- ICEYE SAR: X-band spotlight mode at approximately 0.5–1 m resolution. Revisit at a given site can reach daily with tasking coordination. Provides the primary overcast-period fallback alongside Capella.
Why aggregate rather than sum individual tanks
A single floating-roof tank yields a fill-level estimate with a measurement uncertainty of roughly ±1–3 percentage points under good conditions, driven by sun-angle error, image geolocation, and the finite precision of shadow-edge detection at sub-metre resolution. For a trader or analyst who needs a site-level inventory figure, that per-tank error is the starting point, not the answer.
When a facility holds 40 or 60 tanks, aggregation changes the arithmetic entirely. If per-tank errors are statistically independent, the standard error of the site total scales as the per-tank uncertainty divided by the square root of the number of tanks. Forty tanks with ±2% individual uncertainty yield a site-level standard error closer to ±0.3% of total capacity, assuming errors are uncorrelated. In practice they are not fully independent: shared sun angle, shared image acquisition time, and shared atmospheric conditions introduce positive correlation. A realistic effective sample size is often 60–70% of the actual tank count, which still compresses uncertainty substantially. The practical benefit of aggregation is that meaningful injection or withdrawal signals, on the order of 2–5% of site capacity, become detectable even when no single tank tells a clean story.
The leg-landing problem: when physics defeats geometry
Floating roofs do not float indefinitely. When a tank is drawn down to roughly 10–15% of capacity, the roof settles onto its support legs, typically a ring of fixed steel columns. Below that threshold, the roof position is decoupled from the liquid level. Shadow geometry then reports a stable apparent fill level regardless of how much product remains, or whether the tank is being cleaned, inspected, or refilled from the bottom.
This ambiguity is not detectable from shadow geometry alone. Analysts must flag any tank whose shadow implies a fill level below the leg-landing threshold, typically 10–20% depending on tank design, and exclude it from the aggregate or treat it as a bounded range rather than a point estimate. At a large terminal, 5–10% of tanks may be in this state at any given time. Excluding them slightly reduces the effective sample size but avoids systematic bias. SAR backscatter from the roof deck offers a partial cross-check: a roof resting on legs tends to show a different backscatter pattern than one floating freely, though interpreting this reliably requires site-specific calibration.
SAR fills the gap when cloud closes the optical window
Optical shadow analysis fails completely under cloud cover. For facilities in the North Sea basin, the Gulf of Mexico coast, or equatorial refining hubs, overcast periods can persist for days or weeks. Capella Space and ICEYE both operate X-band SAR in spotlight mode at approximately 0.5 m resolution, sufficient to resolve shadow geometry on tanks above roughly 15 m diameter.
SAR shadow analysis follows the same geometric principle as optical: the roof casts a radar shadow whose length on the tank floor is a function of the sensor depression angle and the roof height above the floor. The depression angle for SAR is typically 20–50 degrees from horizontal, well-documented in each satellite's published imaging geometry. The principal complication is that SAR shadow edges are less sharp than optical shadows due to speckle, requiring multi-look averaging or filtering that slightly degrades effective resolution. In practice, SAR-derived fill levels on large tanks carry an additional ±2–3 percentage points of uncertainty compared with optical, but the aggregate benefit still applies. The combination of Capella or ICEYE on overcast days with Pléiades Neo or SkySat on clear days allows near-continuous site monitoring with a latency of 24–48 hours in most cases.
One further SAR diagnostic is roof-deck coherence. Between two SAR acquisitions separated by days, a floating roof that has moved vertically produces low interferometric coherence at its location. A stationary roof on its legs produces high coherence. This provides a qualitative injection or withdrawal signal independent of shadow geometry, though converting it to a volume estimate requires additional modelling.
Uncertainty propagation and what the output actually says
A site-level inventory estimate is not a meter reading. It is a probability distribution over stored volume, and the honest output names that distribution explicitly. The key parameters are: per-tank measurement uncertainty (typically ±1–3% of individual capacity for optical, ±3–5% for SAR); the inter-tank correlation coefficient (empirically 0.2–0.4 for same-pass imagery); the fraction of tanks excluded due to leg-landing ambiguity; and the tank capacity register, which must come from public records or client data and carries its own uncertainty if nominal capacities are used.
Change detection between two epochs is often more useful than absolute inventory. If the same methodology is applied consistently across two acquisitions, systematic biases partially cancel. A site that showed an aggregate shadow-implied fill of 58% in one pass and 63% in the next, with a combined uncertainty of ±1.5 percentage points on the change, is a meaningful signal even if the absolute 58% figure carries ±4% uncertainty. Communicating this distinction to commodity analysts and risk teams is as important as the number itself.
Building a monitoring cadence for a specific terminal
The practical design of a monitoring programme depends on the terminal's latitude, typical cloud frequency, tank count, and the minimum detectable change the client needs. A Gulf Coast terminal with 60 tanks and a 5% change threshold can be served adequately by twice-weekly optical tasking supplemented by SAR on cloudy days. A terminal in a persistently overcast region may need SAR as the primary sensor with optical as the occasional validation pass.
Archive depth matters for context. Planet's SkySat archive extends back to around 2017 for many locations; Pléiades tasking archives from 2012. Capella launched in 2020 and ICEYE's commercial archive from 2019. For establishing a baseline seasonal pattern of injection and withdrawal, combining available historical optical imagery with current tasking is standard practice. Satellize applies this multi-source aggregation methodology operationally, including in commodity-intelligence contexts, and the same statistical framework used in the Tonga crop-estimation programme for uncertainty propagation across spatial samples transfers directly to tank-farm aggregation problems.
What the method cannot do
Shadow analysis measures roof position, not product type or quality. A tank nominally holding crude may hold condensate, fuel oil, or water during maintenance. The method cannot distinguish these without supplementary information. It also cannot see inside fixed-roof tanks, which are common at older facilities and at product terminals. Sites with a mix of floating and fixed-roof tanks will have a partial observability problem that must be documented in any output.
Very small tanks, below roughly 15 m diameter, produce shadows too narrow to measure reliably at 0.5 m resolution once geolocation error and shadow-edge uncertainty are accounted for. Facilities dominated by small tanks are poor candidates for this method. Finally, the method produces an inventory estimate, not a flow rate. Inferring injection or withdrawal rates requires at least two acquisitions with a known time interval, and the rate uncertainty is the quadrature sum of the two epoch uncertainties divided by the time difference. For short intervals between passes, rate uncertainty can be large enough to obscure all but the most dramatic movements.
Typical figures
| Optical spatial resolution | 30 cm (Pléiades Neo); 50 cm (SkySat) |
| SAR spatial resolution | ~0.5 m spotlight mode (Capella X-band, ICEYE X-band) |
| Minimum tank diameter for reliable shadow measurement | ~15 m (SAR); ~10 m (optical at 30 cm) |
| Typical revisit with tasking | Daily to twice-weekly; SAR fills optical gaps during cloud |
| Per-tank fill-level uncertainty | ±1–3% of capacity (optical); ±3–5% (SAR) |
| Site-level aggregate uncertainty (40+ tanks) | ±0.5–2% of total site capacity, depending on inter-tank correlation |
| Leg-landing exclusion zone | Tanks implied below ~10–20% fill excluded or flagged as bounded range |
| Archive depth | Optical from ~2012 (Pléiades); SAR from ~2019–2020 (ICEYE, Capella) |
| Delivery latency | 24–48 hours from acquisition for tasked passes |
| Sensor frequency (SAR) | X-band (~9.6 GHz); all-weather, day/night capable |
Analytics Satellize can run
| Site-level inventory estimate with uncertainty bounds | Shadow-geometry aggregation with correlated-error propagation across tank population | Structured report: total estimated volume, confidence interval, excluded-tank log, per-epoch timestamp |
| Inventory change signal between two epochs | Differenced aggregate fill estimates; systematic bias cancellation; rate uncertainty calculation | Change report with net injection or withdrawal estimate in barrels and as percentage of capacity |
| Leg-landing flag layer | Per-tank fill-level threshold check against published or estimated leg height; SAR backscatter cross-check where available | GIS layer marking each tank as measurable, leg-landed, or ambiguous at each epoch |
| SAR-derived roof movement coherence map | Interferometric coherence between repeat SAR passes; low coherence indicates vertical roof displacement | Qualitative injection or withdrawal signal map per pass pair, flagging tanks with significant movement |
| Monitoring cadence design and cloud-risk assessment | Historical cloud-frequency analysis from Sentinel-2 scene metadata or ERA5 cloud cover climatology for the facility location | Recommended sensor mix and tasking schedule; expected data-gap probability by month |
| Historical baseline inventory time series | Retrospective shadow analysis on archived optical and SAR imagery | Time-series dataset of site-level fill estimates from archive depth to present, in tabular and GIS formats |
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.