Mine access road traffic density and haulage route monitoring
Very-high-resolution optical imagery from constellations such as WorldView Legion and Pléiades Neo can detect and count individual haul trucks, giving analysts a production-correlated traffic signal that sits entirely outside company reporting chains.
Sensors
- Maxar WorldView Legion: Panchromatic resolution of 30 cm, multispectral at 1.2 m. Up to 15 revisits per day over a target site when the full six-satellite constellation is operational, making it the highest-cadence sub-metre commercial option for traffic counting.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral, four-satellite constellation. Revisit of 1–2 days at mid-latitudes under clear skies; stereo and tri-stereo modes available for road-surface DEM extraction alongside truck detection.
- Planet SkySat: 50 cm panchromatic resolution, tasked on demand. Revisit is operator-scheduled rather than orbital, typically 1–3 passes per day over a given site. Useful for spot-checks and validating anomalies flagged by coarser sensors.
- Sentinel-1 SAR (C-band): 10 m ground range detected resolution in IW mode, 6-day repeat (12-day for a single satellite). Coherence change detection reveals road surface disturbance from heavy traffic even through cloud and at night, though it cannot resolve individual trucks.
Why haul roads are the mine's most legible signal
A large open-pit mine is a system of constrained flows. Ore and waste move from the pit face to the crusher, the stockpile or the dump along a fixed network of haul roads. That network does not change week to week. The trucks that use it are enormous: a Caterpillar 797F has a body width of roughly 9.1 metres. At 30 cm resolution, a loaded 797F is a 30-pixel-wide object with a shadow that extends further still. It is not hard to find.
The value of counting those objects is that haulage volume is a direct upstream proxy for mine output. Operators know this; it is why shift supervisors count truck cycles. What changes when a satellite does the counting is that the count becomes available to anyone with access to the imagery, independent of what the operator chooses to disclose. For investors assessing production guidance, regulators checking royalty declarations, or supply-chain teams verifying offtake volumes, that independence is the entire point.
What sub-metre imagery actually resolves, and what it does not
At 30 cm panchromatic resolution, individual large haul trucks are detectable with high confidence. Smaller light vehicles, pickup trucks and personnel carriers are detectable but harder to classify reliably without multispectral context or shadow geometry. Articulated road trains on access corridors present as elongated bright objects and can be distinguished from rigid-body trucks by length-to-width ratio.
Detection confidence drops sharply in shadow zones on the pit walls, under dust plumes, and wherever road surfaces share spectral characteristics with the vehicles. Cloud cover is a hard blocker for optical sensors. A site in a tropical or monsoonal climate may lose 60–80 per cent of acquisition opportunities to cloud in the wet season, which is precisely when some operators prefer to move material. SAR coherence on Sentinel-1 partially compensates: heavy traffic degrades road-surface coherence in ways that persist in the 6-day interferometric pair, giving a qualitative busy/quiet signal even when optical imagery is unavailable. It will not give you a truck count.
Revisit frequency is the binding constraint for anything resembling a continuous traffic census. Even WorldView Legion at up to 15 passes per day over a target does not provide the kind of continuous monitoring a fixed camera would. What it provides is a statistically useful sample, from which daily or weekly traffic-density estimates can be constructed if the sampling is frequent enough and the mine's operating rhythm is understood.
Building a traffic count from a sample of snapshots
The analytical method is straightforward in principle. Object detection, typically a convolutional neural network trained on labelled VHR imagery of mining vehicles, runs on each acquired scene. Detected vehicles are georeferenced and assigned to road segments. Counts per segment per acquisition are aggregated into a time series. From that time series, analysts derive metrics: average vehicles per hour on a given road segment, directional flow ratios (loaded trucks heading to the crusher versus empty trucks returning to the pit face), and anomaly flags when counts deviate significantly from the site's established baseline.
The honest caveat is that a snapshot count is not a throughput figure. To convert truck counts to tonnes moved, you need payload assumptions. Published payload ratings for common ultra-class trucks range from 290 to 400 tonnes, but actual payloads vary with material density and loading practice. Analysts typically apply a payload range rather than a point estimate, which produces a production-volume band rather than a single number. That band is still useful for trend analysis and for detecting step-changes in activity.
SAR coherence as a weather-independent corroborating layer
Sentinel-1 Interferometric Wide Swath mode produces 10 m resolution imagery on a 6-day repeat cycle, freely available through the Copernicus Data Space. When a road surface is repeatedly disturbed by heavy vehicle traffic between two acquisition dates, the phase coherence between the two SAR images is reduced. Stable, unused surfaces retain high coherence. The result is a coherence map that highlights active versus inactive road segments without needing to see a single truck.
This is particularly useful for access corridors that run through dense forest canopy, where optical sensors see treetops rather than road surfaces. SAR penetrates the canopy partially at C-band, and the road surface, being a specular reflector, often shows up in backscatter even under tree cover. The method cannot count trucks or estimate tonnage, but it can confirm whether a road is being used at all, and it works in rain.
Operational limits buyers should price in
Latency between image acquisition and a verified truck count depends on tasking lead time, cloud-free acquisition, and processing pipeline speed. For a site with daily WorldView Legion tasking under clear skies, a same-day or next-day count is achievable. For a cloud-affected tropical site, gaps of one to three weeks between usable optical acquisitions are realistic in the wet season.
Archive depth matters for baseline construction. Maxar's archive extends to 2008 for WorldView-1 and earlier sensors, though sub-30 cm imagery is more recent. Planet's SkySat archive begins around 2014. Pléiades Neo launched in 2021. For a mine that has been operating for decades, older Landsat and SPOT imagery can establish a coarser historical context, but individual truck detection requires imagery acquired after the VHR commercial constellations became available.
Satellize runs vehicle-detection analytics on tasked VHR imagery and can combine optical counts with Sentinel-1 coherence layers into a unified haulage-activity index. The approach is the same one underlying our crop-estimation work in Tonga: build a repeatable, sensor-agnostic pipeline that produces a number with honest uncertainty bounds, not a black-box output.
What a monitoring programme looks like in practice
A practical haulage monitoring programme for a single large open-pit site typically involves tasking one or two VHR satellites on a daily or near-daily basis, accepting that cloud will reduce effective acquisition frequency. Sentinel-1 coherence runs automatically on the 6-day repeat. Object-detection results feed a road-segment database that accumulates a rolling 90-day traffic history.
Deliverables are usually a weekly summary report with per-segment truck counts, a directional flow chart, a production-volume estimate with uncertainty range, and an alert if counts drop or spike by more than two standard deviations from the trailing 30-day mean. GIS layers in GeoTIFF or vector format allow clients to overlay counts on their own mine-plan data. The archive is the asset: after six months of monitoring, the baseline is solid enough that anomaly detection becomes genuinely useful rather than noise-dominated.
Typical figures
| Best optical resolution (panchromatic) | 30 cm (WorldView Legion, Pléiades Neo) |
| Best optical resolution (multispectral) | 1.2 m (WorldView Legion, Pléiades Neo) |
| Maximum revisit (optical, tasked) | Up to 15 passes per day over a target (WorldView Legion full constellation) |
| SAR resolution and revisit | 10 m (Sentinel-1 IW mode); 6-day repeat per satellite, 12-day single-satellite |
| Minimum detectable target | Ultra-class haul trucks (body width ~9 m) detectable with high confidence at 30 cm; light vehicles (width ~2 m) detectable but classification less reliable |
| Cloud impact | Hard blocker for optical; SAR unaffected. Tropical wet-season optical availability can fall below 20–40% of acquisition attempts |
| Archive depth (VHR optical) | Pléiades Neo from 2021; SkySat from ~2014; Maxar archive from 2008 (coarser sensors earlier) |
| Typical latency (optical count to delivery) | Same-day to next-day under clear skies with daily tasking; 1–3 weeks during persistent cloud cover |
| Delivery formats | GeoTIFF detection layers, vector road-segment shapefiles, CSV time series, PDF weekly summary report |
Analytics Satellize can run
| Per-segment truck count time series | Convolutional neural network object detection on VHR panchromatic imagery, georeferenced to road-segment network | CSV or GIS vector layer with counts per segment per acquisition date, updated on each clear-sky pass |
| Directional flow ratio (loaded vs. empty) | Shadow geometry and vehicle spacing analysis to infer direction of travel on one-way haul roads | Weekly chart showing loaded-to-empty truck ratio per road segment, flagging imbalances |
| Production-volume estimate with uncertainty band | Truck count multiplied by published payload-rating range for detected vehicle class; Monte Carlo uncertainty propagation | Weekly production-volume range (tonnes) in summary report, with stated assumptions |
| Traffic anomaly alert | Statistical process control on rolling 30-day truck-count baseline; alert triggered at two standard deviations | Email or API alert with annotated scene thumbnail and count deviation magnitude |
| Road-activity coherence map (SAR) | Sentinel-1 interferometric coherence change detection between 6-day acquisition pairs | GeoTIFF coherence-difference layer showing active vs. inactive road segments, delivered fortnightly |
| Haulage-activity index | Fusion of optical truck counts and SAR coherence into a normalised weekly index, gap-filled using coherence during cloud periods | Time-series index chart and underlying data table, suitable for trend comparison against reported production figures |
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.