Drone Stockpile Surveys in Mining: RTK Without Infrastructure

Volumetrics you can defend: repeatable centimetre surveys at pits where CORS has never reached.

Why stockpiles are the perfect drone job — and the perfect trap

Stockpile volumetrics were drone mapping's first killer application: a 20-hectare yard that took a rover crew a day of climbing becomes a 25-minute flight, with denser surfaces and nobody walking on unstable material. A Matrice 350 RTK with an L2 LiDAR or P1 camera, or a Mavic 3 Enterprise on smaller yards, produces surfaces that reconcile with weighbridge tickets to within a couple of percent.

The trap is longitudinal. Volumes are differences between surfaces captured weeks apart, and differencing amplifies any misalignment between epochs. A base position that wanders between visits — the signature of averaged single-point setups — tilts and shifts each month's surface slightly, and thousands of cubic metres of phantom gain or loss appear across a large pad. Month-over-month numbers that swing inexplicably while each individual survey looks perfect: that is the relative-accuracy trap wearing a hi-vis vest (unpacked fully in absolute vs relative accuracy).

The infrastructure reality at mine sites

Mines are where correction infrastructure goes to die. CORS networks rarely reach leases hours from the nearest town; cellular coverage stops at the gate or the pit rim; and survey monuments have a hard life around blasting, haul traffic, and rehabilitation earthworks. Crews have historically answered with static sessions posted to OPUS or AUSPOS (accurate, but next-day), site-established control networks (expensive to maintain), or metre-class averaged bases (the trap above). Network RTK subscriptions solve nothing where there is no network.

This is precisely the environment a self-converging PPP base was built for: L-Band corrections arrive by satellite regardless of remoteness, the base computes its own ~1.5 cm ITRF2020 position in about three minutes, and the pit gets centimetre RTK with zero dependence on town (remote-area patterns here).

A monthly workflow that survives audits

Repeatable process beats heroics. Base placement: same general area each visit — the rim bench with open sky — though the exact spot no longer matters because coordinates are absolute, not monument-inherited. Converge, start base mode, enable RINEX logging. Fly the standard mission plan (same altitude, overlap, and boundary each month makes surfaces statistically comparable). Confirm FIX in DJI Pilot 2 throughout; any leg that dropped to FLOAT behind a highwall gets post-processed from the base log afterwards. Before demobilizing, shoot two checkpoints on stable ground — a concrete plinth, a paint mark on the ROM pad — and file the residuals with the survey.

Processing follows the same discipline: identical photogrammetry parameters, surfaces exported in ITRF2020 (or the mine grid via one documented transformation), volumes computed pad-by-pad against the previous epoch. When the auditor asks why April gained 3,000 m³, the answer is in the data trail, not in anyone's memory.

Accuracy budget for volumetrics, in numbers

Volume error scales with surface error times footprint area. A 2 cm vertical bias across a 200 m × 100 m pad is 400 m³ — often inside contract tolerance; a 15 cm bias from a FLOAT-flown epoch is 3,000 m³ and a difficult phone call. Centimetre absolute control of both epochs keeps the differencing honest; consistent flight parameters keep photogrammetric noise symmetric so it cancels in the subtraction; and checkpoints quantify what remains. Teams that publish a standing accuracy statement — base method, checkpoint residuals, processing versions — find that disputes with contractors and regulators mostly stop happening.

Beyond stockpiles: the same base, the rest of the mine

The correction stream that anchors volumetrics is site-wide capital. Haul-road grading with machine control consumes the identical RTCM 3.x broadcast; pit-wall monitoring rovers, tailings-dam inspection flights, and rehabilitation as-builts all inherit the same absolute datum, so every dataset the site produces overlays every other by construction (the machine-control pattern). One base at the rim, radio across the lease, NTRIP where the office network reaches: mines that consolidate on a single correction source typically retire a patchwork of subscriptions, hand-me-down bases, and datum-mismatch folklore in the process.

Practical notes from the field

Dust and weather: an IP68 shell earns its rating on a mine; wipe the antenna dome occasionally and it simply works. Blasting schedules: converge and fly outside exclusion windows; the base powers up and re-converges in minutes after an all-clear, so clearing the pad costs almost nothing. Night shifts: the 10-hour battery covers a shift, and USB-C external power covers a campaign. Data custody: RINEX logs and checkpoint files go to the survey database the same day — the habit that turns any future coordinate question into a five-minute lookup rather than a re-survey.

Reconciling with the weighbridge: closing the loop

Volumetrics earn trust when they reconcile with independent numbers, and mines have the best independent number in industry: the weighbridge. The reconciliation habit is straightforward — convert survey volume deltas to tonnes via measured product density, compare against loadout tickets for the same period, and track the residual percentage month over month. Well-run programs sit inside 1–2% on formed piles; a residual that suddenly jumps flags either a density change (moisture, product blend) or a survey problem, and the absolute datum makes the diagnosis fast because the geometric side is beyond suspicion. Auditors respond to this loop viscerally: a twelve-month chart of survey-versus-weighbridge residuals is worth more than any accuracy certificate, because it is the accuracy claim being tested twelve times.

Getting started on a working mine: a 30-day plan

Week one: place and converge the base, establish the two permanent checkpoints on stable concrete, fly the baseline survey of every active pad, and file the first checkpoint residuals. Week two: repeat one pad mid-week to measure your own survey-to-survey noise floor — the number every future volume delta is judged against. Week three: fold in the monthly pads, standardize the flight plans, and hand the console credentials to the second pilot (single-person dependencies die on rosters). Week four: run the first weighbridge reconciliation and present the residual chart. By day 30 the program has a defensible noise floor, an audit trail, and no dependence on town infrastructure — which is more than many legacy programs establish in a year.

One-line takeaway

Absolute base coordinates make monthly surfaces subtractable, RINEX logging makes every flight rescuable, and a weighbridge reconciliation chart makes the whole program audit-proof — the rest of stockpile accuracy is flight discipline.

Further reading

The accuracy theory behind subtractable surfaces is in absolute vs relative accuracy; the offline correction patterns for leases beyond coverage are in remote-area RTK.

Frequently asked questions

How accurate are drone stockpile volumes really?

With centimetre absolute control of both epochs and consistent flight parameters, 1–2% agreement with weighbridge reconciliation is routine on well-formed piles; irregular, vegetated, or trafficked piles degrade gracefully from there.

Do I need GCPs on every stockpile flight?

With an absolutely-positioned base and RTK geotags, most sites keep 2–3 permanent checkpoints for verification instead of laying GCP grids — see the GCP guide.

What if the pit has no cellular at all?

Nothing changes: corrections reach the base by L-Band satellite and reach the drone by UHF radio or RC link. The workflow is designed for exactly that site.

Can the same setup serve LiDAR and photogrammetry?

Yes — the base doesn't care about the payload. L2 LiDAR strips and P1 imagery both consume the same RTK stream and land on the same datum.