Mine production activity inference from truck and equipment counts
High-revisit optical and SAR imagery lets analysts count haul trucks and excavators to estimate ore throughput before any disclosure. Production slowdowns, surges and idle periods become visible days or weeks ahead of official reporting.
Sensors
- Planet SkySat: 0.5 m native resolution optical; task-on-demand revisit of once or twice daily over a target site. Resolves individual haul trucks (typically 10–15 m long) as distinct objects and supports frame-by-frame change detection within a single day.
- Maxar WorldView-3: 0.31 m panchromatic, 1.24 m multispectral. At this resolution truck cab colour, shadow geometry and tyre-track freshness are all readable. Revisit is roughly 1–4.5 days depending on latitude and tasking priority; archive depth extends to 2014.
- Capella Space SAR (X-band): Spotlight mode delivers roughly 0.35–1 m resolution. X-band SAR is unaffected by the thick dust plumes that routinely saturate optical sensors over active pits, and it images at night. Haul trucks appear as bright point scatterers against the low-backscatter pit floor, making counts feasible even in zero-visibility optical conditions.
- Planet PlanetScope: 3 m resolution, near-daily global coverage from a constellation of over 180 Dove satellites. Too coarse for individual truck identification, but conveyor-belt activity, stockpile shadow change and overall pit-floor reflectance shifts are detectable at this cadence and usable for daily tempo scoring.
- Sentinel-2 MSI: 10 m optical at 10–12 day revisit (5 days with both satellites). Insufficient for truck counts but useful as a free, consistent baseline for site-level activity indices: bare-soil disturbance extent, stockpile volume proxies and haul-road freshness.
Why truck counts are a proxy for ore throughput
Open-pit mining is a materials-handling problem at its heart. The number of loaded haul-truck cycles per shift is directly proportional to the volume of ore and waste moved. At large copper or iron-ore operations, a standard fleet of 300-tonne class trucks running at capacity can move 150,000–200,000 tonnes per day. When that fleet sits idle, or when only a fraction of trucks are moving, throughput falls in step. No corporate communication is required to see it.
Satellite analysts exploit this by counting trucks visible in sequential images, noting their position on haul roads versus loading zones, and tracking whether conveyors cast moving shadows or remain static. The correlation between visible truck activity and reported quarterly production has been studied in the context of commodity-market intelligence; the relationship is not perfect, but directional signals of ±15–20% throughput change are generally detectable from imagery alone. That is enough to matter when a mine accounts for a meaningful share of global supply of a specific metal.
What the physics of dust actually does to your data
Active open pits generate continuous dust plumes from blasting, truck movement and dry ore handling. On optical imagery, a thick plume can obscure the pit floor entirely, making truck counts impossible for that pass. This is not a rare edge case; at arid-climate mines in the Atacama, Pilbara or Central Asian copper belts, dust is a near-daily problem during dry seasons.
X-band SAR cuts through it. Dust particles are orders of magnitude smaller than the 3.1 cm wavelength of X-band radar, so the signal passes through unattenuated and reflects off the metal bodies of trucks and excavators. Capella's spotlight mode at sub-metre resolution produces bright, distinct returns from haul trucks even when the pit floor is optically invisible. The practical workflow pairs morning optical tasking from SkySat or WorldView-3 with SAR passes timed for dust-peak afternoon or night hours, giving two independent counts per day at high-priority sites.
Object detection: what the models can and cannot do
Convolutional neural networks trained on labelled mining-equipment imagery can detect and classify haul trucks, rope shovels and front-end loaders in sub-metre optical imagery with published detection rates in the 85–95% range under clear conditions, based on peer-reviewed work in the remote sensing literature. False positives from static equipment, parked vehicles and rock shadows are the main error source. Shadow geometry filtering and multi-temporal consistency checks reduce but do not eliminate them.
SAR object detection is harder. Trucks appear as bright point scatterers, but so do metal structures, cable anchors and survey markers. Without an optical reference image to co-register against, classification confidence drops. The honest position: SAR counts are best treated as corroborating evidence rather than primary counts, particularly at mines with dense fixed infrastructure. Resolution also matters sharply. At 1 m SAR resolution a 15 m truck is a handful of pixels; at 0.35 m it becomes a recognisable elongated scatterer. Capella's highest-resolution spotlight mode is therefore worth the tasking premium for this application.
Conveyor activity adds a third signal that does not require object detection at all. A moving conveyor belt produces a distinctive linear reflectance signature and, in time-series stacks, a repeatable pattern of stockpile growth that can be modelled geometrically from shadow length changes. It is a slower signal than truck counts but is visible even in 3–10 m imagery.
Building a production tempo index
A single truck count on a single day is almost meaningless without context. The useful product is a normalised activity index built from a rolling time series: trucks observed divided by the estimated maximum fleet size, averaged over a trailing window of seven to thirty days. Fleet size is estimated from peak-count days in the historical archive, cross-referenced with public permit filings or equipment procurement announcements where available.
The index runs from zero (no visible activity) to one (apparent full capacity). Sustained readings below 0.4 for more than a week are a credible signal of unplanned downtime, labour action or equipment failure. Readings above 0.9 sustained for several weeks can indicate a production push ahead of a price window or a contractual deadline. Neither interpretation is certain from imagery alone. Metallurgical recoveries, ore grade variation and haulage distance all affect the throughput implied by a given truck count, and none of those are observable from orbit.
Satellize runs this kind of activity-index methodology on open constellations and commercial tasking under client licence. The Kingdom of Tonga crop-estimation programme uses a structurally similar approach, building a normalised activity index from time-series imagery rather than counting equipment, which gives a sense of how the same analytical pattern applies across sectors.
Honest limits: what imagery cannot tell you
Ore grade is invisible. Two mines with identical truck counts can have wildly different metal output if one is processing high-grade ore and the other is moving waste. Imagery tells you how hard a mine is working, not how productively.
Underground operations are largely opaque. Surface indicators, including shaft-head activity, ore-car movement and stockpile change, provide partial signals, but the bulk of production from block-cave or underground longwall operations happens out of sight. This method is most reliable at open-pit operations, which account for the majority of global copper, iron ore and gold production but not all of it.
Cloud cover remains a hard constraint for optical sensors. Tropical and equatorial mines, including many in the DRC, Indonesia and Papua New Guinea, can be cloud-obscured for days at a time. SAR partially compensates, but the combined optical-plus-SAR revisit still has gaps. Buyers should expect that perhaps 10–25% of scheduled optical passes over cloud-prone sites will return unusable imagery in any given month, depending on season.
From image to intelligence: the delivery chain
Raw imagery is not an intelligence product. The useful output is a structured alert or periodic report that states: on this date, at this site, truck count was X, representing Y% of estimated fleet capacity, up or down Z% from the trailing 30-day average. That framing gives a commodity analyst, an investor or a government regulator something actionable without requiring them to interpret imagery themselves.
Latency from image acquisition to delivered alert is typically 4–24 hours for commercial tasking pipelines, depending on processing queue and downlink schedule. Archive analysis, going back through historical imagery to reconstruct a production history for a site, can cover years at Maxar or Planet archive depth. That retrospective capability is often more valuable than the real-time feed: it lets a buyer understand the baseline before committing to ongoing monitoring.
Typical figures
| Best optical resolution | 0.31 m (Maxar WorldView-3 panchromatic); 0.5 m (Planet SkySat) |
| Best SAR resolution | ~0.35 m (Capella Space spotlight mode, X-band) |
| Optical revisit at target site | Daily or twice-daily with SkySat tasking; 1–4.5 days with WorldView-3 |
| SAR revisit | Capella: multiple passes per day possible with tasking; independent of solar illumination and dust |
| Minimum detectable object | Haul trucks (~10–15 m body length) reliably detected at ≤0.5 m optical or ≤0.5 m SAR; marginal at 1 m |
| Spectral bands (optical) | Panchromatic + 8-band VNIR/SWIR (WorldView-3); RGB + NIR (SkySat, PlanetScope) |
| SAR frequency | X-band (~9.6 GHz, ~3.1 cm wavelength) for Capella; cloud and dust penetrating |
| Archive depth | Maxar: 2008 onwards; Planet: 2016 onwards at useful resolution; Sentinel-2: 2015 onwards (free) |
| Alert latency | 4–24 hours from acquisition to structured report for commercial tasking pipelines |
| Delivery formats | GeoTIFF chip extracts, JSON activity-index feed, PDF periodic report, GIS vector layer (truck positions) |
Analytics Satellize can run
| Daily truck count and fleet utilisation index | CNN-based object detection on sub-metre optical imagery; point-scatterer detection on SAR; normalised against estimated peak fleet size from historical archive | Structured JSON alert with count, utilisation percentage and delta from 30-day trailing average |
| Production tempo time series | Rolling activity index computed from multi-sensor time-series stack; gap-filled using SAR on cloud or dust days | Monthly PDF report with annotated time-series chart and flagged anomaly dates |
| Conveyor activity indicator | Linear feature detection and shadow-length change modelling on 3–10 m imagery; stockpile footprint measurement | GIS polygon layer with conveyor-active / conveyor-idle classification per pass |
| Unplanned downtime alert | Threshold trigger when 7-day rolling utilisation index drops below configurable floor (default 0.4) relative to site baseline | Email or API push alert within 24 hours of threshold breach, with supporting image chip |
| Retrospective production history reconstruction | Archive analysis across Maxar and Planet historical catalogues; manual QA on ambiguous frames; fleet-size estimation from peak-count days | Tabular dataset of monthly activity indices covering available archive depth, delivered as CSV and PDF |
| Multi-site comparative dashboard | Standardised activity index applied across a portfolio of named sites; normalised for pit geometry and fleet class differences | Web-hosted dashboard with site-level sparklines, anomaly flags and downloadable data export |
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.