Grain silo and open-stockpile monitoring for commodity intelligence
Satellite optical and SAR data can track grain inventories at silos, port terminals, and open storage pads, giving commodity analysts an independent read on stock levels before official figures are published.
Sensors
- Planet SkySat: 50 cm native resolution (resampled product at 50 cm), tasked revisit typically same-day to next-day. Sufficient to resolve individual silo shadow lengths and measure open-pile footprints for volume estimation. Limited to cloud-free conditions.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral across six bands including red-edge and near-infrared. Stereo and tri-stereo tasking enables photogrammetric digital surface models of open piles; accuracy typically within 0.3 m in height under good geometry. Revisit up to twice daily at mid-latitudes.
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral, plus eight SWIR bands at 3.7 m. The SWIR bands (1195–2365 nm) allow spectral discrimination between grain, coal, and mineral ores, which is not possible with visible bands alone. Tasked revisit typically one to four days depending on latitude and priority.
- Sentinel-1 SAR (C-band, ESA): IW mode ground range resolution approximately 5 × 20 m, free and open, six-day repeat at the equator (three days with both satellites). C-band backscatter responds to surface roughness and dielectric properties. Grain piles produce characteristic backscatter signatures distinct from smooth concrete or water; useful for detecting large fill/draw events even through cloud.
- Sentinel-2 MSI (ESA): 10 m visible and near-infrared bands, 20 m red-edge and SWIR, five-day revisit at the equator. Too coarse to resolve individual silos but useful for mapping the spectral signature of open grain pads at port scale and for regional harvest-pulse context that explains why inventories are moving.
What a silo shadow actually tells you
Covered cylindrical silos are structurally analogous to floating-roof oil tanks, with one important difference: the roof does not move with the contents. What changes is the apparent height of grain visible above the silo rim, which is rarely detectable directly. The useful signal instead comes from the shadow cast by the silo cylinder itself onto surrounding ground. Shadow length is a function of solar elevation angle, silo height, and the height of any grain mounded above the rim. When a silo is filled to overflowing, grain crowns above the lip and extends the effective casting height by one to three metres. That extension is measurable at 30–50 cm resolution with consistent solar geometry.
The method requires precise knowledge of the silo's structural height, which can be extracted from a baseline stereo acquisition or from publicly available facility records. Subsequent single-pass images then become inventory snapshots. Accuracy degrades when shadows fall on uneven ground or overlap adjacent structures, and the technique is blind to partially filled silos where grain sits below the rim. Honest expectation: this approach gives a binary high/low signal more reliably than a precise fill percentage.
Open piles are a cleaner geometry problem
Open storage pads, common at port terminals and inland receival points during harvest, offer a more tractable measurement problem. A grain pile approximates a cone or truncated cone. Its volume is a function of base area and height, both of which stereo photogrammetry can recover. Pléiades Neo tri-stereo acquisitions routinely produce digital surface models with 0.5 m posting and vertical accuracy in the 0.2–0.5 m range under good sun angle and low haze. For a pile 10 m high and 60 m in diameter, that translates to volume uncertainty of roughly 5–10 per cent, which is competitive with many ground-based measurement methods.
Converting volume to mass requires bulk density, which varies by grain type, moisture content, and compaction. Wheat typically runs 750–800 kg/m³; maize around 720 kg/m³; soybeans around 720–740 kg/m³. These ranges are well established in agricultural engineering literature. The practical consequence is that satellite-derived volumes carry an additional 3–5 per cent mass uncertainty from density assumptions alone, and analysts should communicate this clearly rather than presenting tonne figures as precise.
Telling grain from coal from ore: the spectral argument
At visible wavelengths, a large pile of yellow wheat and a pile of pale limestone can look similar in a panchromatic image. Discrimination matters because port yards often store multiple commodities on adjacent pads, and misidentification corrupts inventory estimates. WorldView-3's eight SWIR bands are the most useful tool here. Grain has a characteristic spectral response in the 900–1300 nm range driven by starch and protein absorption features; coal is strongly absorbing across SWIR; mineral ores vary but generally lack the organic absorption features that grain exhibits.
Sentinel-2's 20 m SWIR bands (bands 11 and 12, centred at 1610 nm and 2190 nm) provide a coarser but free alternative for large pads. The limitation is spatial: at 20 m, a pad must be at least 40–60 m across before the spectral reading is uncontaminated by surrounding concrete or soil. For most major port terminals this threshold is met; for small inland silos it rarely is. SAR backscatter adds a complementary dimension. Grain surfaces are rough at C-band wavelengths (5.6 cm), producing moderate to high backscatter, whereas smooth concrete returns very little. This helps delineate pile extents in cloud-affected imagery when optical data is unavailable.
Harvest pulses and the problem of rapid turnover
Grain storage is not static infrastructure. A terminal that is full in March can be empty by May as vessels load out. A silo that appears empty in a single image may have turned over twice in the interval between acquisitions. This makes revisit rate the binding constraint, not resolution. A weekly optical cadence, achievable with SkySat tasking over a priority site, catches most significant movements. A monthly cadence misses them entirely.
Seasonal context matters too. Inventory levels at a Black Sea terminal in October, immediately post-harvest, are not comparable to levels in June. Regional crop-condition data from Sentinel-2 NDVI time series provides the harvest-pulse context that makes a single inventory snapshot interpretable. Without it, a full silo in October is unremarkable; a full silo in June, in a drought year, is a signal worth acting on. Satellize's crop-estimation work for the Kingdom of Tonga uses exactly this kind of phenological context to distinguish structural from seasonal variation.
Archive depth and the baseline problem
Any change-detection approach needs a baseline. For grain facilities, the most useful baseline is not a single image but a seasonal stack covering at least two to three years, sufficient to establish typical fill cycles for a given terminal. The Sentinel-1 archive extends back to 2014; Sentinel-2 to 2015; Planet's archive for many sites to 2016 or earlier. Commercial tasking archives for Pléiades and WorldView-3 are thinner and less consistent but can be supplemented with declassified or commercial archive purchases.
A practical programme for a new site typically begins with a three-month archive pull to establish baseline geometry and spectral signatures, followed by weekly or fortnightly tasking during the commodity's active trading season. Off-season monitoring can drop to monthly without significant loss of intelligence, which reduces cost considerably. The key structural parameters, silo heights, pad dimensions, conveyor routing, are extracted once and reused across all subsequent acquisitions.
Honest limits of the method
Cloud is the most persistent constraint. Tropical and temperate coastal terminals can lose weeks of optical coverage during monsoon or winter storm seasons. SAR fills part of the gap but cannot replace stereo volume estimation; it confirms presence and rough extent, not precise volume. Snow cover on open piles in northern-hemisphere winter creates a false surface that inflates height measurements and must be flagged or excluded.
Resolution floors matter at smaller facilities. A country elevator with 10 m diameter silos is at the edge of what SkySat can resolve for shadow analysis, and below what Sentinel-2 can characterise spectrally. The method is most reliable at large export terminals and major inland hubs where pad dimensions and silo clusters are large enough to provide clean measurements. Analysts should resist applying port-scale confidence intervals to smaller inland facilities without validation.
Typical figures
| Best optical resolution (tasked) | 30 cm panchromatic (Pléiades Neo, WorldView-3) |
| Open-pile height accuracy (stereo) | 0.2–0.5 m vertical under good sun angle and low haze (Pléiades Neo tri-stereo) |
| SAR resolution (Sentinel-1 IW mode) | ~5 × 20 m ground range; free and open |
| Optical revisit (tasked) | Same-day to next-day (SkySat); up to twice daily (Pléiades Neo) |
| SAR revisit (Sentinel-1, two satellites) | 6 days at equator; 3 days with both Sentinel-1A and 1B |
| Spectral bands for commodity discrimination | 8 SWIR bands at 3.7 m (WorldView-3, 1195–2365 nm); Sentinel-2 bands 11 and 12 at 20 m |
| Minimum detectable pile (stereo volume method) | Approximately 40–60 m pad diameter for reliable spectral characterisation at Sentinel-2 resolution |
| Archive depth | Sentinel-1 from 2014; Sentinel-2 from 2015; Planet from ~2016; commercial archives variable |
| Volume-to-mass density uncertainty | 3–5% additional uncertainty from bulk density assumptions (grain type and moisture dependent) |
| Delivery formats | GeoTIFF DSM, GeoJSON pile polygons, CSV inventory time series, PDF intelligence report |
Analytics Satellize can run
| Silo fill-level change signal | Shadow-length analysis from high-resolution optical imagery using known structural height as baseline; binary high/low classification with change flagging | Weekly change alert with confidence flag, delivered as GeoJSON attribute layer |
| Open-pile volume estimate | Stereo photogrammetric digital surface model differenced against bare-pad baseline; cone/frustum volume integration | Per-pile volume in cubic metres with uncertainty range; GeoTIFF DSM and CSV summary |
| Commodity type classification | SWIR spectral signature matching against grain, coal, and mineral ore reference libraries using WorldView-3 or Sentinel-2 bands | Classified pad map as GeoJSON with commodity label and confidence score |
| SAR-based fill/draw event detection | Sentinel-1 C-band backscatter time series over known storage pads; anomaly detection against seasonal baseline | Bi-weekly event log with backscatter change magnitude; cloud-resilient complement to optical series |
| Harvest-pulse inventory context | Sentinel-2 NDVI phenological time series for surrounding agricultural catchment correlated with terminal inventory observations | Seasonal inventory narrative report with NDVI anomaly map and inventory trend chart |
| Facility baseline geometry extraction | One-time stereo acquisition to extract silo heights, pad dimensions, and conveyor layout; stored as persistent site model | Site geometry GeoJSON and annotated facility map; reused across all subsequent monitoring acquisitions |
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.