Palm oil mill throughput estimation from SAR-detected fresh fruit bunch stockpiles
C-band SAR backscatter distinguishes fresh fruit bunch heaps from bare yard surfaces, letting analysts track stockpile extent across repeat passes and infer crude palm oil throughput without setting foot on site.
Sensors
- Sentinel-1 C-band SAR (ESA): 10 m ground range resolution in Interferometric Wide Swath mode, 6-day repeat at mid-latitudes with both satellites operational, 12-day with one. Penetrates cloud and operates at night, critical for equatorial mill yards. Free and open archive from 2014.
- COSMO-SkyMed (ASI): X-band SAR at 1-3 m resolution in Spotlight mode. Shorter wavelength increases sensitivity to surface texture differences between FFB heaps and concrete, useful for smaller yards where Sentinel-1 pixels blend targets. Tasked commercially.
- ICEYE SAR constellation: X-band, sub-1 m resolution in Spotlight mode, with same-day or next-day revisit possible through commercial tasking. Useful for time-sensitive throughput snapshots around quarterly reporting dates.
- Planet SuperDove optical: 3 m resolution, near-daily revisit. Cannot see through cloud, but on clear days provides colour confirmation of pile extent and helps distinguish FFB from other yard materials such as empty fruit bunches or palm kernel shells, which SAR alone cannot separate reliably.
Why a pile of fruit is legible to radar
Fresh fruit bunches are awkward cargo: spiky, dense, high in moisture, and typically stacked in irregular heaps one to three metres tall in open mill yards before sterilisation. Those physical properties make them unusually visible in C-band SAR imagery. The rough, wet surface of an FFB heap produces strong diffuse backscatter, typically several decibels above the return from a flat concrete or compacted laterite yard. Sentinel-1's C-band (5.405 GHz, roughly 5.5 cm wavelength) is well matched to the surface roughness scale of FFB heaps, which have feature sizes in the centimetre-to-decimetre range.
The practical consequence is that a well-calibrated Sentinel-1 scene can distinguish an active stockpile from an empty yard, and changes in the bright-return footprint between consecutive passes track changes in pile area. This is not a subtle signal buried in noise. It is one of the cleaner backscatter contrasts available in agricultural SAR work, comparable in clarity to flooded rice paddies.
From pile area to tonnes of CPO: the inference chain
Area is observable. Volume is not, at least not from a single-pass SAR image without repeat-pass interferometric height retrieval, which requires coherent pairs and is rarely achievable over loose, shifting agricultural material. The height ambiguity is real and should not be papered over. In practice, analysts work with a pile-area proxy calibrated against ground-truth measurements or optical stereo where available, and apply a mean-height assumption derived from mill-type priors. A typical FFB heap at a medium-scale mill runs one to two metres in depth; at large-scale mills with front-end loaders, piles can reach three metres.
Once an FFB volume estimate is in hand, conversion to crude palm oil uses the oil extraction rate (OER), which for commercially operated mills in Malaysia and Indonesia typically falls between 19 and 23 percent by weight of FFB processed, as documented in industry and government statistics. The throughput inference is therefore: pile-volume change over the inter-pass interval, multiplied by assumed bulk density of FFB (roughly 600-700 kg per cubic metre), multiplied by OER. Each assumption carries uncertainty, and honest reporting should propagate that uncertainty rather than present a single point estimate.
What the method cannot see, and what partially compensates
The limits are worth naming plainly. Pile height is the dominant unknown. Single-pass SAR gives area, not volume. Repeat-pass InSAR over FFB heaps is theoretically possible but coherence is poor over loose organic material that shifts between passes. Optical stereo from commercial satellites (Pleiades Neo, WorldView Legion) can recover height at better than 50 cm vertical accuracy in good conditions, and fusing stereo-derived height with SAR-derived area extent is the most defensible volume method available without ground access.
Processing-rate priors are the second uncertainty. Mill OER varies with fruit ripeness, mill age, maintenance quality, and whether the mill is running at full capacity. A mill operating at 60 percent capacity has a different throughput per unit of FFB stockpile than one running flat out. Analysts can partially constrain this by tracking the rate of pile drawdown between passes: a pile that shrinks quickly suggests active processing; one that grows over multiple passes suggests either a harvest surge or a maintenance shutdown. Neither inference is unambiguous without corroborating data.
Cloud cover is not an issue for SAR, which is its principal advantage over optical methods in equatorial regions where cloud fraction exceeds 70 percent on average. However, very heavy tropical rainfall can temporarily increase backscatter from yard surfaces, potentially inflating apparent pile extent. This is a known artefact in C-band agricultural SAR and should be flagged in any operational system.
Who uses this, and why the timing matters
The primary buyers are commodity traders and analysts tracking Indonesian and Malaysian CPO production ahead of monthly supply-demand reports, agricultural lenders assessing mill collateral, and ESG-focused investors who want independent verification of declared throughput figures. Palm oil is one of the most traded vegetable oils globally, and production estimates from government agencies in Indonesia and Malaysia are published with a lag of several weeks. A satellite-derived nowcast can sharpen a position or flag a discrepancy weeks before official data appears.
Insurance applications are narrower but real. A mill that declares a production loss following an equipment failure or flood event can be assessed against the SAR-derived pile accumulation record. If stockpiles built up without drawdown during the claimed outage period, that is material information for a loss adjuster. The archive depth of Sentinel-1, running continuously since 2014 for Sentinel-1A and 2016 for Sentinel-1B (with Sentinel-1C launched in 2024), means that multi-year baseline behaviour for individual mills is already available without any new tasking cost.
Putting numbers on the detection floor
At Sentinel-1's 10 m resolution in IW mode, a pile occupying fewer than two or three pixels is difficult to separate reliably from speckle. That corresponds roughly to a heap covering less than 200-300 square metres, which at a mean height of 1.5 m and FFB bulk density of 650 kg per cubic metre represents around 200-300 tonnes of FFB. Smaller mills processing under 20 tonnes of FFB per hour may have stockpiles near or below this detection floor during off-peak periods. For those sites, COSMO-SkyMed or ICEYE at 1-3 m resolution extends detection to piles of a few tens of tonnes.
Satellize runs this analysis on Sentinel-1 open-archive data for regional mill-network monitoring, with commercial SAR tasking added for specific mills where clients require higher temporal or spatial precision. The approach is structurally similar to the crop-estimation work done for the Kingdom of Tonga programme, where known agronomic priors are combined with satellite observables to produce calibrated quantity estimates rather than raw index values.
Typical figures
| Primary SAR spatial resolution | 10 m (Sentinel-1 IW mode); 1-3 m (COSMO-SkyMed / ICEYE Spotlight) |
| Revisit interval | 6 days (Sentinel-1 dual-satellite); 1-2 days commercially tasked (ICEYE) |
| Radar frequency / band | C-band 5.405 GHz (Sentinel-1); X-band ~9.6 GHz (COSMO-SkyMed, ICEYE) |
| Minimum detectable pile area (Sentinel-1) | Approximately 200-300 m² (2-3 pixels); smaller with X-band systems |
| Approximate FFB mass detection floor (Sentinel-1) | Roughly 200-300 tonnes FFB at mean 1.5 m pile height; lower with stereo height fusion |
| Cloud sensitivity | None for SAR; optical confirmation (Planet SuperDove, 3 m) cloud-limited |
| Archive depth | Sentinel-1A from April 2014; Sentinel-1B from April 2016; Sentinel-1C from 2024 |
| Analysis latency after acquisition | Sentinel-1 NRT products available within 1-3 hours of pass; analysis turnaround typically same day |
| Throughput conversion parameter (OER) | 19-23% by weight of FFB, varies by mill and fruit quality (industry-published range) |
| Delivery formats | GeoTIFF pile-extent polygons, CSV throughput time series, PDF mill-level report |
Analytics Satellize can run
| Mill-yard stockpile extent map | Calibrated C-band backscatter thresholding and change detection on multi-temporal Sentinel-1 IW GRD scenes | GeoTIFF polygon layer per pass date, with area in square metres and backscatter intensity statistics per pile |
| FFB volume time series | Area-to-volume conversion using mill-type height priors; optional fusion with optical stereo DSM where available | CSV time series of estimated FFB volume (m³) and mass (tonnes) per mill, with stated uncertainty range |
| CPO throughput nowcast | Volume drawdown rate multiplied by oil extraction rate prior (19-23% OER range), propagated with Monte Carlo uncertainty | Weekly PDF mill report with point estimate and 80% confidence interval for CPO output in tonnes |
| Processing-status classification | Pile growth/drawdown rate classification: active processing, accumulation, shutdown, or ambiguous | Per-pass status flag in structured JSON feed, suitable for integration into commodity analytics platforms |
| Declared-versus-observed throughput flag | Comparison of satellite-derived throughput estimate against mill-declared or government-reported figures, with statistical significance test | Discrepancy alert with supporting imagery and methodology note, formatted for loss-adjustment or ESG audit use |
| Multi-year baseline mill profile | Seasonal decomposition of Sentinel-1 archive backscatter time series (2014 to present) to establish normal operating envelope per mill | Historical benchmark report per mill, showing typical seasonal throughput pattern and anomaly thresholds |
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.