Can Satellites See All of China's Oil Storage?
Satellite gauging is the backbone of every public estimate of Chinese crude inventories, and it is a partial instrument by design. It reads the roofs of floating roof tanks well. It cannot read fixed roof tanks, and it cannot read underground rock caverns at all. The gap is not a technical accident, it is a storage design choice with a published cost advantage.
- Satellite gauging works by measuring the vertical position of a floating roof, which rests directly on the liquid. The Baker Institute analysis describes computer vision that can deduce inventory levels by analysing the movements of floating oil storage tank roofs over time. That is the whole mechanism, and it only exists on one tank type.
- Fixed roof tanks defeat the method. Their rigid lid obstructs both visual and radar based remote analysis. The Baker Institute work puts the construction premium at 10 to 20 per cent over a floating roof tank, which is a modest price for removing an asset from the observable set.
- Underground storage removes the asset entirely. Water sealed rock caverns cannot be gauged from orbit by any sensor, and cost over 60 per cent less per barrel than aboveground tanks according to the same analysis. China has completed at least one subterranean site with more under construction, which is covered in our note on whether China stores oil underground.
- The economics of concealment are cheap at the margin and expensive in bulk. Doming a single 300 foot diameter tank is put at 1.0 to 1.2 million US dollars. Concealing the entire 52 tank Zhoushan facility is estimated at around 60 million US dollars, a rounding error against a reserve worth tens of billions.
- The historical precedent for how large the error can be is substantial. Orbital Insight satellite data indicated Chinese crude stockpiles in mid to late 2017 were more than three times the figures reported by Xinhua, a discrepancy exceeding 500 million barrels. Treat that as a historical illustration of method divergence, not as a current estimate.
- Watching costs money and the bill is public. Monthly high resolution monitoring of key Chinese storage infrastructure is estimated at roughly 4.5 million US dollars a year, or about 1.1 million for quarterly coverage. That cost is one reason coverage is uneven between providers and between months.
It measures a lid, not a liquid
There is a persistent misreading of satellite oil intelligence, which is that an orbiting sensor somehow sees inside a tank. It does not. The entire method rests on a mechanical accident of one tank design. An external floating roof tank has no fixed lid. Its roof is a buoyant deck that sits directly on the surface of the crude and rises and falls with the volume beneath it. Measure the height of that deck against the known height and diameter of the shell, and you have a volume.
Two sensor families do the measuring. Optical imagery reads the shadow that the tank wall casts onto the sunken roof, and the length of that shadow against solar geometry gives the roof depth. Synthetic aperture radar reads the same geometry without needing daylight or clear sky, which matters over coastal China where cloud cover is routine. The European Space Agency's Copernicus programme has documented Sentinel-1 radar being used to monitor oil industry activity on exactly this principle.
The commercial layer above those sensors is crowded and has been for a decade. The Baker Institute analysis names Orbital Insight, Genscape, Planet Labs, DigitalGlobe, ICEYE, Ursa Space Systems, Airbus and TankerTrackers as participants in this market. The names most often attached to Chinese crude numbers in trade press today are Kpler, Vortexa, Kayrros and Ursa Space Systems. They are not measuring different things. They are applying the same roof geometry method to different tank samples, with different revisit rates and different interpolation when a tank is missed.
This is why the correct question is never whether a provider is accurate. It is which tanks are in that provider's panel, and what the panel is assumed to represent.
Project 54The only storage mode an orbiting sensor can gauge is the one with a roof that floats on the oil. Everything else in the estimate is inference.Two storage modes sit outside the method, and both are cheaper
The first blind spot is the fixed roof tank. A fixed roof tank has a permanent rigid lid welded above the liquid, with a vapour space between. There is no moving surface to measure and no shadow that encodes a fill level. The Baker Institute work states plainly that these tanks obstruct both visual and radar based remote analysis. The cost of adopting them is given as 10 to 20 per cent above a floating roof tank of the same capacity.
The second blind spot is underground. China's strategic programme combines coastal steel tank farms with water sealed rock caverns, a design already covered in our note on underground storage. A cavern is rock with oil in it. It emits no thermal or radar signature that distinguishes a full chamber from an empty one, and there is no roof to watch. The Baker Institute analysis puts mined cavern storage at over 60 per cent cheaper per barrel than aboveground tanks, and records at least one completed Chinese subterranean site with several more under construction.
Put those two together and the incentive structure becomes obvious. The modes that remove an asset from satellite observation are not the expensive modes. One costs a small premium, the other is substantially cheaper than the alternative it replaces. A reserve operator optimising only on cost would still drift toward the unobservable half of the estimate. Opacity is a by product of a rational build decision, which makes it durable in a way that a policy of secrecy would not be.
A third limitation is more mundane and gets less attention. Coverage is bought. Monthly high resolution monitoring of key Chinese storage infrastructure is costed at approximately 4.5 million US dollars annually, falling to about 1.1 million for quarterly monitoring. Providers make commercial choices about which sites justify that spend, which means the observable panel is itself shaped by budget, not only by physics.
About 60 million US dollars to blind a 52 tank facility
The Baker Institute analysis prices concealment directly, which is unusually useful because it converts an intelligence question into a capital question. Adding protective doming to a single tank of 300 foot diameter is put at 1.0 to 1.2 million US dollars. Extending that to the full 52 tank facility at Zhoushan is estimated at around 60 million US dollars.
Set that against the asset being concealed. At a crude price in the region of 65 to 75 US dollars a barrel, the inventory in a facility of that scale is worth billions. A 60 million US dollar spend to remove that inventory from the public record is not a deterrent, it is a rounding error. The same logic applies with more force to new build decisions, where choosing fixed roof or cavern construction costs nothing extra in the cavern case and very little in the fixed roof case.
The practical conclusion for anyone relying on satellite derived Chinese inventory data is that the observable share is a policy variable, not a constant. If the visible fraction of Chinese storage falls over time, published estimates will show inventories flattening even if real inventories are rising, because the denominator of the sample is shrinking while the extrapolation assumptions stay fixed. That failure mode is silent. Nothing in the data flags it.
| Storage mode | Satellite readable | Relative cost | What the sensor sees |
|---|---|---|---|
| External floating roof tank | Yes, this is the method | Baseline | Roof height against shell, by optical shadow or radar |
| Fixed roof tank | No | 10 to 20 per cent more than floating roof | A rigid lid, with no volume information beneath it |
| Domed or retrofitted tank | No | 1.0 to 1.2 million US dollars per 300 foot tank | A cover, fitted after construction |
| Underground water sealed cavern | No | Over 60 per cent cheaper per barrel | Nothing. There is no surface expression of fill |
The supply balance, which has a different and equally real error
The standard counterweight to satellite gauging is the supply balance, sometimes called the implied stock build. It is arithmetic rather than observation. Take crude imports plus domestic production, subtract refinery throughput, and treat the residual as a change in inventory. It has one decisive advantage over satellites, which is that it is blind to storage mode. A barrel that goes into a rock cavern shows up in the residual exactly as a barrel that goes into a floating roof tank.
Its weakness is on the other side of the equation. The subtraction depends on knowing refinery runs, and a large and variable share of Chinese throughput sits with independent refiners whose utilisation is itself estimated rather than reported. An error in assumed run rates propagates straight into the implied build at full size. The method also cannot distinguish commercial from strategic inventory, because it only produces a national net number.
This is why the two methods disagree, and why the disagreement is informative rather than embarrassing. They fail in opposite directions. Satellites undercount by excluding unobservable modes. The supply balance captures everything but allocates it badly and inherits refinery estimation error. Where the two converge, confidence is reasonable. Where they diverge, the gap is a rough indicator of how much crude has moved into storage the sensors cannot reach. Our longer treatment of why the estimates disagree sets out the full reconciliation.
For a practitioner the honest framing is that China's crude inventory is a bounded range, not a figure. Total capacity sits somewhere around 1.8 to 2.4 billion barrels on published third party estimates, with actual fill commonly placed near 1.2 to 1.5 billion barrels. The only official capacity number China has ever released remains 238 million barrels across nine strategic bases, published by the National Bureau of Statistics in 2017.
Stop quoting a number, start quoting a method and a bound
Chinese inventory figures circulate through energy sales and marketing material constantly, usually as a bare number with no provenance, because a bare number is easier to put on a slide. That habit is a credibility risk in front of a technical buyer who knows the measurement is partial. The recoverable position is to lead with the method.
Three disciplines follow. First, always state which method produced the figure you are citing, satellite gauging or supply balance, because the two are not interchangeable and a reader who knows the field will test you on it. Second, quote a range and name the sources that bound it, rather than a point estimate borrowed from a headline. Third, say explicitly that the estimate excludes underground and fixed roof storage when it is satellite derived, which is the single sentence that converts a weak claim into a defensible one.
The broader lesson generalises past China. Any commercial argument that rests on remotely sensed data inherits the sensor's blind spots, and the party you are selling to is increasingly likely to know what those are. Precision that is not earned reads as carelessness. A stated bound with a named method reads as competence, and it survives the follow up question, which the point estimate never does.
Listen & take it with you
Prefer audio, or need the deck for an internal review? The full briefing is available as a podcast episode and a downloadable slide presentation.
When you cite a Chinese crude inventory figure, what do you attach to it?
Frequently asked
No. Satellite gauging reads external floating roof tanks, where the roof rests on the crude and its height gives a volume. Fixed roof tanks block the method because a rigid lid carries no fill information, and underground water sealed rock caverns produce no surface signature at all. Published satellite derived totals for China therefore describe the visible subset of storage, not the whole.
By measuring the vertical position of a floating roof. Optical sensors read the shadow the tank wall casts onto the sunken roof and convert shadow length into roof depth using solar geometry. Synthetic aperture radar measures the same geometry without daylight or clear sky, which matters over cloudy coastal regions. The European Space Agency has documented Sentinel-1 radar being used for oil industry monitoring on this principle.
Because there is no moving surface to observe and no volume signature that reaches the surface. A water sealed rock cavern is a mined chamber holding crude below ground, and a full chamber looks the same from orbit as an empty one. The Baker Institute analysis by Gabriel Collins and Shih Yu (Elsie) Hung puts mined cavern storage at over 60 per cent cheaper per barrel than aboveground tanks, so the unobservable mode is also the cheaper one.
The Baker Institute analysis estimates roughly 1.0 to 1.2 million US dollars to add protective doming to a single tank of 300 foot diameter, and around 60 million US dollars to conceal the full 52 tank facility at Zhoushan. Building fixed roof rather than floating roof tanks achieves a similar result for a construction premium of only 10 to 20 per cent.
The Baker Institute analysis names Orbital Insight, Genscape, Planet Labs, DigitalGlobe, ICEYE, Ursa Space Systems, Airbus and TankerTrackers. In current trade coverage of Chinese crude the names most often cited are Kpler, Vortexa, Kayrros and Ursa Space Systems. They apply the same roof geometry method to different tank panels with different revisit rates, so differences between them reflect sample and timing rather than a different measurement principle.
Get the next intelligence drop
Join energy and industrial leaders getting our marketing, AI-growth and revenue-architecture intelligence, direct, no filler.
You're on the list
Welcome to The Energy Growth Brief, watch your inbox for the next dispatch.