Grain silo and bulk-storage inventory estimation by SAR
Synthetic aperture radar can estimate the volume of grain stockpiles and covered silos through cloud cover, day or night, giving commodity traders and lenders an independent physical check on declared inventory.
Sensors
- Sentinel-1 C-band SAR (ESA): 5 x 20 m resolution in Interferometric Wide Swath mode; 6-12 day revisit at mid-latitudes; freely available archive from 2014. C-band (5.405 GHz) penetrates light cloud and is sensitive to large metallic structures via double-bounce, but moisture on grain surfaces significantly raises backscatter and can mimic a larger pile.
- ICEYE X-band SAR: Spotlight mode delivers sub-1 m resolution; X-band (9.65 GHz) produces sharper shadow boundaries than C-band, improving height retrieval for smaller piles. Tasked commercially; revisit to a specific site can be daily with constellation scheduling.
- Capella Space X-band SAR: Spotlight mode at approximately 0.35 m resolution; fine enough to resolve individual silo roof deformation and distinguish adjacent stockpile cones. Tasked commercially with typical delivery latency of a few hours after acquisition.
- Umbra X-band SAR: Spotlight mode at up to 0.25 m resolution, among the finest commercially available. At this resolution, shadow length measurement for a 5 m tall pile becomes geometrically tractable. Coherent change detection between repeat passes can flag material movement between visits.
What a floating roof gives away, and what a pile of grain cannot hide
Commodity inventory fraud depends on an auditor accepting a number without visiting the physical asset. Satellite SAR removes that option. A conical grain stockpile sitting on a concrete pad returns a characteristic signature: a bright double-bounce return at the pile base where the near-vertical grain face meets the horizontal pad, and a radar shadow extending downrange whose length is a direct function of pile height given a known incidence angle. The geometry is elementary trigonometry, not inference.
Covered cylindrical silos present a different problem. The roof is opaque to optical sensors but the silo wall height, diameter and roof curvature are measurable in high-resolution SAR imagery through layover and shadow analysis. A steel silo also produces a strong specular return from its curved wall that varies predictably with incidence angle. Cross-referencing wall height against declared capacity (which is a function of internal volume, itself a function of diameter and height) gives a physical upper bound on what the facility can hold, regardless of what any warehouse receipt says.
The physics of SAR shadow-based height retrieval
In a side-looking SAR image, a vertical object of height H at incidence angle θ casts a radar shadow of length S = H / tan(θ) in slant-range geometry, or H × cos(θ) / sin(θ) in ground-range projection. Sentinel-1 IW mode typically operates between 29° and 46° incidence; at 35°, a 10 m pile casts a ground-range shadow of roughly 14 m, which is resolvable at 20 m pixel spacing only marginally. At X-band sub-metre resolution, the same shadow spans tens of pixels and height uncertainty drops to roughly 0.5 to 1 m for a well-shaped pile.
Volume estimation requires a shape assumption. Grain stockpiles are commonly modelled as cones (angle of repose for dry wheat is approximately 25-28°) or as truncated cones where the base has been partially reclaimed. The cone assumption introduces volume error of perhaps 10-20% when the actual pile has been partially loaded or drawn down asymmetrically. This is not a fatal limitation for a lender checking whether declared inventory is plausible, but it matters for a trader trying to estimate tonnes to the nearest hundred. Repeat-pass differencing, comparing shadow geometry across two acquisitions, is more reliable for detecting material movement than for computing absolute volume.
Where the method struggles: moisture, shape and minimum pile size
Moisture is the most consequential confound. Wet grain or a rained-on stockpile surface raises C-band backscatter substantially, because the dielectric constant of water is roughly 80 versus 3-5 for dry grain. A wet pile can appear brighter and apparently larger in backscatter intensity. Shadow geometry is immune to this effect, which is one reason shadow-based height retrieval is preferred over backscatter-intensity-based volume proxies. X-band is more sensitive to surface moisture than C-band because the shorter wavelength interacts with a shallower surface layer.
Minimum detectable pile height is governed by resolution and shadow contrast. At Sentinel-1's 20 m ground range resolution, a pile shorter than roughly 8-10 m produces a shadow that is one pixel or less, making reliable height retrieval impractical. ICEYE or Capella X-band spotlight imagery at sub-1 m resolution lowers this floor to approximately 2-3 m, which covers most commercially significant stockpiles. Silos shorter than about 5 m with diameters under 10 m approach the detectability limit even at X-band. Small on-farm bins, common in many developing markets, are effectively invisible to current commercial SAR at any band.
Covered storage: reading a silo without opening it
A cylindrical steel silo is a measurable object. Its external diameter is readable in sub-metre SAR or optical imagery; its wall height is recoverable from shadow analysis; its roof type (flat, conical, domed) changes the volume calculation by a few percent. Published engineering standards for grain silos (such as those from the Food and Agriculture Organisation) give standard fill factors and bulk density ranges by crop type, allowing a conversion from geometric volume to estimated tonne capacity. Wheat bulk density is typically 750-800 kg/m³; maize runs 700-720 kg/m³. These are published figures, not proprietary assumptions.
The honest limit here is that SAR tells you the maximum physical capacity of a silo, not its current fill level. Distinguishing a full silo from an empty one requires either thermal infrared (a full silo has more thermal mass and a different diurnal temperature cycle) or ground-penetrating radar, neither of which is currently available from operational satellite constellations at useful resolution. What SAR can do is flag a discrepancy: if a warehouse receipt claims 50,000 tonnes stored at a facility whose total geometric capacity is 30,000 tonnes, the fraud is visible from orbit.
Building an inventory monitoring programme
A practical programme for a commodity lender typically combines free Sentinel-1 archive data for baseline facility mapping with periodic commercial X-band tasking around key dates: harvest, shipment, loan drawdown and repayment. The Sentinel-1 archive back to 2014 is sufficient to establish a site's historical footprint and to identify facilities that did not exist when a loan was originated.
Satellize structures analytics of this type against open constellations first, adding commercial tasking where resolution or revisit demands it. The methodology is the same class used in the Tonga crop-estimation programme: geometric feature extraction combined with published physical constants, not black-box machine learning. Deliverables are site-level volume estimates with explicit uncertainty bounds, not point estimates that invite false precision. A lender who receives a range of 18,000 to 24,000 tonnes, with the reasoning shown, is better served than one who receives '21,400 tonnes' with no error characterisation.
For a trader monitoring a network of terminals across a country, a change-detection feed flagging sites where shadow geometry has shifted by more than a threshold volume between passes is more operationally useful than a periodic full inventory report. Both products are feasible; the right choice depends on the decision cadence.
Typical figures
| Spatial resolution (Sentinel-1 IW) | 5 × 20 m (range × azimuth), ground range |
| Spatial resolution (ICEYE / Capella / Umbra spotlight) | 0.25 – 1 m, depending on mode and provider |
| Revisit (Sentinel-1, mid-latitudes) | 6 days (single satellite); 6–12 days typical with current constellation |
| Revisit (commercial X-band, tasked) | Daily or sub-daily to a named site with constellation scheduling |
| Radar frequency | C-band: 5.405 GHz (Sentinel-1); X-band: ~9.6 GHz (ICEYE, Capella, Umbra) |
| Minimum detectable pile height | ~8–10 m at C-band 20 m resolution; ~2–3 m at X-band sub-1 m resolution |
| Volume estimation uncertainty (open stockpile, good geometry) | ±10–20% for cone-model assumption; lower with multi-pass differencing |
| Archive depth (Sentinel-1) | From April 2014 (Sentinel-1A launch) |
| Delivery formats | GeoTIFF (shadow/height rasters), GeoJSON (site polygons), CSV (volume estimates with uncertainty), PDF site reports |
| Cloud-cover sensitivity | None: SAR is all-weather, day/night |
Analytics Satellize can run
| Facility baseline map | SAR backscatter segmentation and shadow analysis on Sentinel-1 archive; silo diameter and height extraction from X-band spotlight | GeoJSON layer of all detected storage structures at a named site or region, with geometric dimensions and estimated maximum capacity |
| Stockpile volume estimate (single date) | Shadow-length height retrieval at known SAR incidence angle; cone or truncated-cone volume model with published angle-of-repose priors; bulk density conversion using FAO crop density tables | Site-level volume report with explicit uncertainty range and methodology note, delivered as PDF and structured CSV |
| Inventory change alert | Repeat-pass shadow geometry differencing; coherent change detection (for sub-metre X-band pairs) to flag surface disturbance consistent with loading or reclaim | Alert feed (JSON or email) triggered when estimated volume at a monitored site changes by more than a client-defined threshold between passes |
| Declared-vs-physical capacity discrepancy flag | Geometric volume upper bound from SAR dimensions compared against warehouse receipt or declared inventory figure supplied by client | Discrepancy report flagging sites where declared inventory exceeds physical capacity, with supporting imagery and calculation |
| Multi-site portfolio monitoring | Automated pipeline over Sentinel-1 open archive for baseline; commercial tasking scheduled around loan-covenant dates for high-priority sites | Monthly portfolio summary table with per-site volume estimates, change flags and confidence ratings; GIS layer for client mapping tools |
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.