Iron ore and bauxite stockpile volumetrics from SAR-derived digital surface models

SAR interferometry can resolve the surface of an ore or bauxite stockpile to sub-metre vertical accuracy, letting analysts compute volume change between any two acquisition dates. Converting that volume to mass requires honest accounting of bulk density, moisture, and pile-shape assumptions.

Sensors

TanDEM-X (bistatic X-band SAR pair): Bistatic interferometry with TerraSAR-X delivers digital surface models at 0.25 m range resolution in Spotlight mode; vertical accuracy better than 1 m relative and approaching 0.2 m in controlled conditions over flat industrial surfaces. Tasked acquisitions can be scheduled but are not free; archive coverage of specific terminals is patchy.
COSMO-SkyMed Second Generation (X-band SAR): Spotlight-2 mode reaches 0.35 m ground resolution; repeat-pass InSAR over a stable target such as a concrete stockpile pad can achieve decimetre-level height change detection. Constellation of four satellites gives revisit of 12–24 hours over most latitudes, though interferometric coherence degrades if ore surface is disturbed between passes.
ICEYE SAR constellation (X-band): Spot Extended Area mode delivers roughly 1 m resolution with sub-daily revisit capability at select latitudes. Useful for change detection and coarse volumetrics; vertical precision from repeat-pass InSAR is typically 0.5–2 m depending on baseline geometry, less competitive than TanDEM-X bistatic for precise DSM generation.
Sentinel-1 (C-band SAR, ESA): Free, 6-day repeat at mid-latitudes in IW mode (5 × 20 m resolution). C-band penetrates surface moisture layers differently from X-band, introducing a small but measurable phase bias on wet ore. Useful for monitoring large terminals and detecting gross volume changes, but spatial resolution limits detection of small pile movements.

What a SAR interferogram actually measures over a stockpile

Synthetic aperture radar interferometry compares the phase of two complex SAR images acquired from slightly different orbital positions or times. The phase difference encodes the path-length difference to each resolution cell, which translates directly to surface height once the satellite geometry is known. Over a concrete stockpile pad, the bare-ground baseline can be established from a pre-load acquisition or from the TanDEM-X global DEM (12 m posting, released publicly) and then differenced against a current surface model to isolate pile height.

The key distinction is between a bistatic acquisition, where two satellites image simultaneously and coherence is near-perfect, and repeat-pass interferometry, where surface change between passes decorrelates the signal. Ore and bauxite surfaces are rough and shift as reclaimer equipment works the pile. A bistatic TanDEM-X pair is therefore strongly preferred for a single-epoch DSM; repeat-pass COSMO-SkyMed or ICEYE pairs work best when the pile is undisturbed between acquisitions, typically over a weekend or during a vessel-loading pause.

From surface model to volume: the geometry problem

Once a height raster is in hand, volume integration is conceptually simple: sum the height above baseline across every pixel and multiply by pixel area. In practice, three complications arise. First, the baseline DEM must accurately represent the bare pad surface, including any drainage slope or kerb structure. Errors of even 0.3 m in the baseline propagate linearly into the volume estimate. Second, SAR layover and shadow affect steep pile faces. A conical pile with a 38-degree angle of repose (typical for iron ore fines) will produce layover on the near-range face and shadow on the far-range face at typical incidence angles of 25–45 degrees. The shadowed face must be interpolated, usually by fitting a cone or ridge model to the visible portions.

Third, very large piles at major export terminals, such as those at Port Hedland or Tubarão, can exceed 15 m in height. At that scale, the layover zone is substantial and interpolation uncertainty grows. Published studies using TanDEM-X over industrial sites report volume uncertainties in the range of 3–8 percent for well-constrained piles; shadowed flanks push that toward 10–15 percent for the tallest stockpiles. Stating those bounds honestly matters, because a 5 percent error on a 500,000-tonne pile is 25,000 tonnes, a commercially significant quantity.

The density conversion: where physics meets mineralogy

Volume is not what traders or port operators care about. They need mass. The conversion requires bulk density, and bulk density for ore stockpiles is neither fixed nor easily observed from orbit. Iron ore fines (under 6.3 mm) have a bulk density typically between 1.9 and 2.2 t/m³ depending on particle size distribution and compaction. Lump ore runs slightly lower at 1.6–1.9 t/m³ because of the larger void fraction. Bauxite is lighter still, commonly 1.2–1.5 t/m³, and is far more sensitive to moisture: a bauxite pile that has been rained on for 48 hours can be 10–15 percent heavier per unit volume than a dry pile of the same apparent shape.

Copper concentrate is denser, around 1.8–2.2 t/m³, but is almost always stored under cover precisely because moisture control is critical for shipping safety. Open stockpiles of copper concentrate are therefore less common, and SAR-based volumetrics are correspondingly less applicable.

Moisture content is the dominant source of systematic error in mass estimation. An analyst who applies a single bulk density figure without accounting for recent rainfall will produce biased results. The honest approach is to report volume with explicit uncertainty bounds and to pair the SAR-derived volume with a local precipitation record or a Sentinel-1 backscatter moisture proxy to bracket the density range. The resulting mass estimate then carries a stated uncertainty interval rather than a false point estimate.

Acquisition planning: coherence, incidence angle, and scheduling

Getting a usable interferogram requires planning that goes beyond simply ordering an image. For repeat-pass InSAR, the perpendicular baseline between the two orbits must be large enough to provide height sensitivity but not so large that geometric decorrelation destroys coherence. For X-band systems, the optimal perpendicular baseline for stockpile-scale heights is roughly 200–600 m; TanDEM-X science acquisitions are specifically designed to operate in this range.

Incidence angle selection is a real trade-off. Steep incidence (small angle, say 20–25 degrees) reduces layover on pile faces but compresses the ground range resolution. Shallow incidence (35–45 degrees) improves ground resolution and reduces foreshortening on the pile flanks but increases shadow on the far side. For a terminal where pile orientation is known, the optimal look direction can be computed in advance. Requesting an ascending and a descending pass and merging the two DSMs eliminates most shadow artefacts, at roughly double the cost and scheduling complexity.

What the method cannot do, and what complements it

SAR interferometry measures surface shape, not internal structure. If a pile has been hollowed by reclaimer equipment from below while the outer skin remains intact, the DSM will overestimate volume. This is not a theoretical concern: reclaimer tunnelling is standard practice at some automated terminals. Ground-penetrating radar or load-cell data from the conveyor system are the only ways to catch this discrepancy from outside the terminal fence.

Rain affects SAR in two ways: it attenuates the signal (more so at X-band than C-band) and it changes surface roughness, which alters backscatter intensity and can introduce phase noise. Heavy rainfall events should be flagged in any acquisition log. Wet ore surfaces also shift the apparent scattering centre slightly downward relative to the true surface, a small but non-zero bias.

Satellize runs SAR-based volumetric workflows on commercial X-band tasking combined with Sentinel-1 change detection for inter-acquisition monitoring. The Tonga crop-estimation programme demonstrated the operational pipeline for multi-date raster differencing at scale; the same differencing architecture underpins stockpile DSM change products. For terminals where a client holds a ground-truth weighbridge or draft survey, that data can be used to calibrate the bulk density assumption and reduce mass uncertainty significantly.

Practical accuracy expectations for a procurement decision

A buyer considering this capability needs a realistic accuracy table, not a best-case claim. For a well-illuminated pile with no significant shadow, a TanDEM-X bistatic DSM differenced against a clean baseline will deliver volume uncertainty of roughly 3–6 percent. Apply a bulk density range of plus or minus 0.15 t/m³ and the mass uncertainty widens to 8–12 percent. For a shadowed pile measured by single-pass COSMO-SkyMed, expect 10–15 percent volume uncertainty before density is considered.

Those figures compare favourably with the alternative: a port operator's visual estimate or a periodic manual survey using total station equipment, which is both expensive and infrequent. The satellite method's advantage is not precision alone but cadence. A terminal can be measured after every vessel departure, building a time series that reveals inventory drawdown rates, blending ratios, and discrepancies between reported and observed stock. That time series is where the commercial intelligence lies.

Typical figures

Best achievable ground resolution (X-band Spotlight)	0.25–0.5 m (TanDEM-X / COSMO-SkyMed SG Spotlight-2)
Vertical accuracy, bistatic DSM (TanDEM-X)	0.2–1.0 m relative, depending on incidence angle and surface roughness
Vertical accuracy, repeat-pass InSAR (X-band)	0.5–2.0 m, coherence-dependent; degrades if pile surface is disturbed between passes
Volume estimation uncertainty (well-illuminated pile)	3–8 percent; 10–15 percent for piles with significant SAR shadow
Mass estimation uncertainty (volume + bulk density range)	8–15 percent without ground-truth density calibration
Revisit cadence (ICEYE constellation)	Sub-daily at select latitudes in tasked mode
Revisit cadence (Sentinel-1, free)	6 days at mid-latitudes in IW mode; 12 days near equator on single track
Minimum detectable pile height change	Approximately 0.5 m for X-band bistatic; 1–2 m for C-band repeat-pass
SAR frequency bands used	X-band (9.6 GHz, TanDEM-X, COSMO-SkyMed, ICEYE); C-band (5.4 GHz, Sentinel-1)
Archive depth (Sentinel-1)	From 2014 (Sentinel-1A launch); commercial X-band archives vary by operator

Analytics Satellize can run

Single-epoch digital surface model of stockpile pad	SAR bistatic or repeat-pass interferometry; phase-to-height conversion using published orbital geometry	GeoTIFF height raster at 0.5–2 m posting, with per-pixel coherence mask
Volume estimate per named pile or zone	DSM differencing against bare-ground baseline; cone/ridge model interpolation for shadowed faces	Tabular report: pile ID, estimated volume (m³), shadow-interpolated fraction, volume uncertainty range
Mass estimate with explicit uncertainty interval	Volume multiplied by commodity-specific bulk density range; moisture adjustment using co-located precipitation record or SAR backscatter proxy	Structured data feed: mass low/central/high (tonnes), density assumption, moisture flag
Inventory change time series	Multi-date DSM differencing; change attributed to loading, reclaiming, or blending events using vessel AIS correlation	Monthly or per-vessel-call chart of stock drawdown and replenishment, delivered as PDF report or API feed
Shadow and layover quality mask	Geometric simulation of SAR acquisition geometry over DSM; flagging of pixels where interpolation contributes more than 20 percent of pile volume	GeoTIFF confidence layer included with every DSM delivery
Bulk density calibration model	Regression of SAR-derived volume against client weighbridge or draft-survey mass records; iterative density refinement	Calibrated density lookup table per commodity and season; reduces mass uncertainty to 4–7 percent once at least three ground-truth points are available

Sources

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.