top of page

Port Stockpile Volume: Satellite vs Drone vs Ground Survey

Writer: Srihari Maddula
Srihari Maddula
16 minutes ago
16 min read
A large cargo ship berthed next to a dock
Photo: Efim Borisov on Unsplash

It is the last day of the month at a bulk terminal, and the finance team has three numbers for the same coal yard. The book stock, built from weighbridge records of every rake and truck in and every shipment out, says one figure. The contract surveyor's drone flight last week says a figure about four percent lower. A satellite-based estimate, bought on trial by the commercial team, says something in between. The yard holds tens of thousands of tonnes, so a few percent is a large sum of money, and everyone in the room wants to know which number to believe.

The answer is that they are measuring different things, with different errors, and each is right for a different question. The verdict, stated plainly at the start: use a drone survey with ground control for the numbers that go into the accounts, use satellite stereo for frequent, independent watching of many sites you cannot easily visit, and keep a ground survey or laser scan as the referee for disputes and for calibrating the other two. Free satellite imagery belongs in the picture too, but for the footprint and activity of a yard, not its volume.

The more important finding is about where the error comes from. Resolution gets the attention in sales presentations. In practice the volume error is dominated by three quieter things: the surface you assume under the pile, small systematic height offsets spread over a large footprint, and the conversion from cubic metres to tonnes. A method with excellent resolution and a careless base surface can be worse than a coarser one done properly.

The rest of this post compares the three methods, shows the published accuracy evidence, works through a volume calculation with simulated errors, and ends with a decision table and a monthly reconciliation routine.


Three Ways to Measure a Pile

Every stockpile volume method does the same thing: it measures the top surface of the pile, estimates the ground surface underneath, and integrates the height difference over the pile's footprint. The methods differ in how they measure the top.

Method

How it measures

Typical ground resolution

Vertical accuracy, typical

How often

Needs site access

Drone photogrammetry

Overlapping photographs turned into a 3D surface; ground control points fix scale and height

A few centimetres

A few centimetres with good ground control

Whenever a crew can fly

Yes, plus flight permission

Satellite stereo or tri-stereo

Two or three images of the same place from different angles in one pass

0.3 to 0.7 m

Tenths of a metre

Each tasking, weather permitting

No

Ground survey: GNSS rover or total station

Surveyor walks or sights points on the pile surface and toe

Points, metres apart

Centimetres per point

When scheduled

Yes, and on the pile

Terrestrial laser scanning

Scanner on tripods records dense points from a few stations

Millimetres to centimetres

Millimetres to centimetres

When scheduled

Yes, with line of sight

Free optical and radar imagery

10 m pixels, frequent revisits

10 m

No useful height

Every few days

No

Each row has a role. The mistake is asking one row to answer every question.


What the Published Accuracy Figures Say

Drone stockpile surveys have been tested against conventional methods many times, and the results are consistent: with proper ground control, volumes land within a few percent of the reference, often within one or two.

A study at Clemson University compared drone-based volumetric surveys with traditional survey methods and found that models from nadir images flown at 125 feet ranged from 0.35 to 3.03 percent from the measured values, while oblique images were off by 0.44 to 8.23 percent. Another comparison on a twin gravel stockpile used the volume known from the milling machine as the standard, 2,750 cubic metres; the drone approach gave 2,686 cubic metres and the total station survey 2,831, differences of 2.94 and minus 2.31 percent. Work at an Indonesian coal mine on overburden found that a configuration of six well-distributed ground control points gave a volume difference of 3.76 percent, and recommended at least four uniformly distributed points; badly distributed control produced far larger errors.

Satellite stereo has also matured. An evaluation of very-high-resolution tri-stereo surface models measured vertical accuracy against check points: the Pléiades model had a root-mean-square error of about 0.31 metres, and the WorldView-3 model about 0.13 metres. Researchers have built automatic stockpile monitoring from SkySat multi-view stereo, applied it to large coal stockpiles and validated the volumes against measurements collected on site over the same period.

Source

Method

Reported accuracy

Clemson University study

Drone, nadir images at 125 ft

0.35% to 3.03% from measured volume

Same study

Drone, oblique images

0.44% to 8.23%

Twin gravel stockpile, milling-machine reference

Drone

2.94% from reference

Same

Total station survey

−2.31% from reference

Coal overburden, six ground control points

Drone

3.76% volume difference

Tri-stereo surface model check

Pléiades

About 0.31 m vertical RMSE

Same

WorldView-3

About 0.13 m vertical RMSE

Read the table carefully. The drone figures are volume errors; the satellite figures are height errors. Turning a height error into a volume error depends on the pile, and that is where the next section starts.

Compare methods on volume error for your pile, not on resolution or on a height accuracy quoted for a different site. The conversion between the two is where most claims fall apart.

Why Volume Error Is Mostly About Bias and the Base

Random height noise largely averages out over a pile. A surface model with tens of thousands of cells, each a few centimetres or decimetres off in a random direction, sums to a volume close to the truth. Systematic error does not average out. A surface that sits 10 centimetres too high everywhere adds 10 centimetres times the whole footprint to the volume.

The arithmetic is blunt. For a pile with a footprint of 120 by 60 metres, 7,200 square metres, the volume added by a uniform offset is:

Height offset

Volume added

Share of a pile of about 54,000 m³

0.03 m

216 m³

0.4%

0.10 m

720 m³

1.3%

0.30 m

2,160 m³

4.0%

The base surface, the ground under the pile, behaves the same way. You cannot see it once the pile is there, so you estimate it: from a survey of the empty pad, from the ground around the toe, or from a design level. If the estimate is 10 centimetres low, the volume is 1.3 percent high on this pile, however good the top surface is.

This is why the best practice in every method is the same. Survey the pad when it is empty and keep that surface. Where that is impossible, take the base from the ground immediately around the pile in the same survey, so that any common offset in the survey affects the top and the base equally and cancels. And keep ground control points that tie every survey to the same vertical datum.


The Edge of the Pile and the Noise Floor

There is a second, less obvious source of bias. Volume software usually integrates only the positive heights above the base, because a pile cannot have negative volume. On flat ground around the pile, random noise produces small positive and negative heights; the negative ones are discarded and the positive ones are kept. The noise on the bare ground around the pile therefore adds volume that is not there.

The simulation in this post shows the effect. With zero-mean height noise, integrating over a rectangle that includes a margin of bare ground produced a positive volume error that grew with the noise level. The fix is simple: draw a boundary polygon along the toe of the pile and integrate only inside it, and set the base from the ground just outside that boundary. A survey that reports a volume without saying how the boundary was drawn has left out a number that matters.

Coarse resolution adds its own edge effect. A satellite surface model with half-metre cells blurs the sharp toe of a pile, and small or narrow piles lose a larger share of their volume to blurred edges than large, broad ones. For a yard of many small piles, satellite stereo is weaker than the headline accuracy suggests.


From Cubic Metres to Tonnes: The Density Problem

The finance team does not want cubic metres. It wants tonnes, and the conversion is the largest uncertainty in the whole chain.

Bulk density depends on the material, its size distribution, its moisture and how compacted it is. Coal stacked by a reclaimer, compacted by dozers, rained on and partly dried has a density that varies across a pile and over time. For the pile in this post, a plausible range of densities gives a wide range of tonnes.

Assumed bulk density

Tonnes for 54,082 m³

0.80 t/m³

43,265 t

0.85 t/m³

45,969 t

0.90 t/m³

48,674 t

The densities are illustrative values for stacked coal; your material needs its own. A swing of 0.05 tonnes per cubic metre moves the tonnage by about six percent, more than the volume error of any competent survey method. A dispute about a three percent difference between a drone and a satellite figure is often really a dispute about which density was used.

Agree the density before you argue about the volume. A small change in assumed bulk density moves the tonnage more than the difference between any two competent survey methods.

The remedy is to measure density periodically at each yard for each material, from samples of known volume weighed on calibrated scales, or by reconciling surveyed volumes against weighbridge totals for a pile built from empty. Record the density used with every reported tonnage.


Drone Surveys: Where They Win and Where They Struggle

Drone photogrammetry is the default for accounting-grade stockpile volumes, for good reasons. It is accurate to a few percent or better with proper control, it can be flown at short notice, it covers a whole yard in a single flight, and the surface it produces is detailed enough to show slumps, holes and boundary errors.

Its constraints are practical. A crew needs access to the site and permission to fly. Under India's drone rules, flights must respect the airspace zones shown on the Digital Sky platform, and ports, airports and industrial installations often fall in zones that need permission; plan the approvals before the month-end. Wind, rain and dust affect flight safety and image quality, and coal dust settling on lenses is a real field problem. Shiny, uniform or very dark surfaces give the matching software less texture, and wet coal can be hard. Ground control points must be laid out well and kept: the coal mine study above is a reminder that badly distributed control can wreck accuracy.

The best drone programmes treat ground control as permanent infrastructure: fixed, surveyed markers around each yard that are photographed on every flight, so that every survey sits on the same datum and month-to-month differences are real.


Satellite Stereo: Where It Wins and Where It Struggles

Satellite stereo wins where a drone cannot easily go. It needs no site access, no permit and no crew. One tasking can cover a large yard or several nearby sites at once. Imagery can be bought for past dates from archives, which matters when a dispute is about a stock level three months ago. And it is independent of the site operator, which is the point for lenders, insurers, buyers and regulators who want a view that the operator did not produce.

Its constraints are equally clear. Accuracy is lower: tenths of a metre in height and half-metre cells, which suits large piles and trend monitoring better than small piles and accounting. Each tasking costs money, and the price depends on the provider, the area and the urgency. Cloud blocks optical stereo, which matters in the Indian monsoon. And the processing is specialised: matching, refinement of the camera models and careful base surfaces are what separate a usable result from a misleading one, as the SkySat work above shows.

For a trader or lender watching stock at many ports, the trade is attractive: a consistent, independent volume trend every few weeks across every site, accurate enough to flag a draw-down or an inconsistency, without sending anyone anywhere.


Ground Survey and Laser Scanning: The Referee

Ground methods measure directly. A surveyor with a GNSS rover or a total station records points on the surface and toe; a terrestrial laser scanner records millions of points from a few stations. Both are accurate, both are slow on a large yard, and both put people on or near the piles, which carries safety risk on loose or high material.

Their role in a modern programme is as the referee. A laser scan or a dense ground survey of one pile, done the same day as a drone flight or a satellite pass, calibrates the faster method and settles disputes about it. It is also the method a contract may require for custody transfer, where a certified surveyor's measurement is the legal basis of a transaction.


Free Imagery: Footprint and Activity, Not Volume

Free imagery from Sentinel-2 and Sentinel-1 has 10-metre pixels and no useful height information for stockpiles. It cannot measure volume. It can measure a great deal else, cheaply and often.

From optical imagery every few days you can map the footprint of each pile, see piles appear, grow and disappear, and count how much of a yard is occupied. From radar you can detect activity through cloud: the brightness of a yard changes when material is moved, stacked or reclaimed. Combined with a few volume measurements from a drone or satellite stereo, footprint and activity fill in the weeks between surveys and flag the dates when a new survey is worth buying.

Question

Free imagery answers it?

Is the yard full, half full or empty?

Yes, from footprint area

Was a large pile reclaimed this week?

Yes, from footprint and radar change

How many cubic metres are in pile 7?

No

When should we buy the next stereo pair or fly the drone?

Yes, when the footprint changes


The Monsoon Yard

Most of the published accuracy figures come from dry conditions. An Indian port yard between June and September is a different place, and every method behaves differently in it.

Rain changes the material. Coal and many ores take up water, which raises bulk density and, for some materials, makes piles slump and spread. A pile surveyed before a week of rain and again after it can show less volume and more tonnes, and both readings can be right. If the density used for the second survey is the dry-season figure, the tonnage will be understated, and the book stock will look wrong when it is not.

Rain changes the surface the instruments see. Wet coal is darker and shinier, which gives photogrammetry less texture to match and can leave holes or noise in the drone surface model. Standing water in the yard around the piles confuses the base surface, because water is flat and reflective and matching software struggles with it. Laser scanners see wet surfaces well but lose range in heavy rain.

Rain and cloud change the schedule. Optical satellite stereo needs a clear view, and in the coastal monsoon that can mean weeks without a usable acquisition. Drones cannot fly safely in heavy rain or strong gusts, and a month-end survey day can be lost to weather.

Plan the monsoon months separately: measure density more often, keep a dry-weather window for the month-end flight, and expect satellite stereo to be unavailable for weeks at a time.

A sensible monsoon routine moves the drone survey to the first dry window within a few days of month-end, measures density on wet material at the same time, and relies on free radar imagery, which sees through cloud, to watch footprints and activity while optical sources are blind. The reconciliation then compares like with like: wet tonnes against wet tonnes, with the density that applied on the day.


Choosing by Question

The verdict becomes practical when each question is matched to the cheapest method that answers it well enough.

Question

Best method

Why

Typical frequency

Month-end stock for the accounts

Drone with permanent ground control

Accurate, repeatable, on a stable datum

Monthly

Dispute over a stock figure

Ground survey or laser scan, with agreed density

Direct, defensible, certifiable

When needed

Independent check by a lender or buyer

Satellite stereo

No site access, operator-independent

Monthly or on demand

Watching many remote sites

Free imagery for footprint, stereo for periodic volume

Cheap frequent coverage, volume when it matters

Weekly footprint, monthly volume

Historical stock on a past date

Archived satellite stereo

Only source for the past

Once

Calibrating a drone or satellite programme

Laser scan or dense ground survey of one pile

Reference measurement

Quarterly

Match the method to the question. Accounts want repeatability, disputes want a referee, lenders want independence, and a busy yard wants frequent cheap footprints between the expensive measurements.

The Volume Calculation in Code

The core of every method is the same integration, and the simulation behind the tables above is short. It builds a 120 by 60 metre pile with 35-degree side slopes and a 12-metre crest on flat ground, computes its volume on a fine reference grid, then adds noise and offsets typical of each method and recomputes.

import numpy as np

def pile(cell, L=120.0, W=60.0, H=12.0, margin=20.0):
    nx, ny = int((L + 2 * margin) / cell), int((W + 2 * margin) / cell)
    x = (np.arange(nx) + 0.5) * cell - margin
    y = (np.arange(ny) + 0.5) * cell - margin
    X, Y = np.meshgrid(x, y)
    run = H / np.tan(np.radians(35))                 # horizontal run of the slope
    d = np.minimum(np.minimum(X, L - X), np.minimum(Y, W - Y))
    z = np.clip(d / run, 0, 1) * H
    z[(X < 0) | (X > L) | (Y < 0) | (Y > W)] = 0
    return z

def volume(z, cell, base=0.0):
    # positive heights only; integrate inside a boundary in real use
    return np.clip(z - base, 0, None).sum() * cell * cell

The reference volume came to about 54,082 cubic metres. With 50 noisy repetitions per case, and integrating over the whole rectangle including a 20-metre margin of bare ground, the volume errors were:

Case

Cell size

Height noise

Height offset

Volume error

Drone with ground control

0.05 m

0.03 m

0.02 m

+0.7%

WorldView-3-class tri-stereo

0.5 m

0.13 m

0.03 m

+1.5%

Pléiades-class tri-stereo

0.5 m

0.30 m

0.07 m

+3.5%

Satellite stereo with a poor base

0.5 m

0.30 m

0.30 m

+9.3%

Taking the base from the ground just outside the pile in the same surface model, instead of assuming zero, cut the poor-base case from 9.3 to 2.0 percent, because the common offset cancelled. The remaining error is mostly the noise floor on the bare margin, which a boundary polygon removes. The noise levels and offsets are illustrative choices based on the accuracy figures above, not measurements of any provider.


What a Stockpile Report Should Contain

Most arguments about stockpile figures are arguments about things the report left out. A report that answers the questions below can be checked, compared with the previous month and defended in a dispute.

Item

Why it matters

Example of what to state

Date, time and method

Stock changes by the hour at a busy terminal

Drone survey, 30 Nov, 07:10 to 07:55

Vertical datum and control

Ties this survey to every other one

Eight permanent markers, surveyed by GNSS

Base surface used

The largest single source of systematic error

Empty-pad survey of 12 March, or ring around the toe

Boundary of each pile

Decides where volume is counted

Toe polygons, drawn on the orthomosaic, attached

Volume per pile

The measured quantity

Cubic metres, one decimal place is enough

Density and its source

Converts volume to tonnes; the biggest uncertainty

0.86 t/m³, from samples of 28 Nov

Tonnes per pile and total

The figure finance uses

With the density stated beside it

Estimated uncertainty

Tells the reader how much a difference means

Volume plus or minus 2%, tonnes plus or minus 5%

Change from the previous report

Shows movement and catches errors

Per pile, with in and out totals from the book

The uncertainty line is the one most often missing, and the one that saves most meetings. If the report says the tonnage is good to about five percent once density is included, a four percent gap against the book is not a crisis; it is within the measurement. If the gap is twelve percent, something real has happened, and the report has already ruled out the survey as the cause.


Handling Survey Data at a Port

Ports and terminals are security-sensitive sites, and stockpile data is commercially sensitive. A survey programme should be designed with both in mind from the start.

Drone imagery of a port shows more than piles: berths, vessels, equipment and sometimes restricted areas. Agree with the port authority and the terminal's security team what may be photographed, how long raw images are kept and who may see them, and make the crew's flight logs available. Keep processed products, such as surface models and volumes, separate from raw imagery, and share only the products with people who need volumes.

Stock levels at a trading terminal can move markets for the material concerned. Treat them as confidential business data: limit access, log who downloads reports, and agree in the contract with any survey or satellite provider that the data belongs to the terminal and may not be resold or published. Independent satellite monitoring for a lender or buyer should be covered by the same agreement, with the terminal's knowledge.

Two practical rules keep the programme out of trouble. First, write a retention schedule before the first flight: raw images kept for a short, fixed period once the products are verified, surface models and volumes kept for as long as the accounts require, and everything deleted on schedule rather than when a disk fills up. Second, treat the survey crew as visitors in a hazardous area: inducted, escorted where the terminal requires it, kept off active stacker and reclaimer paths, and never walking on piles to place markers once permanent control is in place. A survey that injures someone or leaks a stock figure costs far more than any accuracy it gained.


A Monthly Reconciliation Routine

A terminal that wants its three numbers to agree can run a simple routine.

  1. Fix the datum. Install permanent, surveyed ground control markers around each yard, and survey each pad when empty to keep its base surface.

  2. Fly at a fixed time. Fly the drone survey on the same day of each month, at a time of day with good light and little activity, and draw boundary polygons along each pile's toe.

  3. Record density. Measure bulk density per material each quarter, or whenever a new source or moisture condition arrives, and attach the density used to every tonnage.

  4. Reconcile against the book. Compare surveyed tonnage with weighbridge book stock per pile and per material, and investigate differences beyond an agreed tolerance rather than averaging them away.

  5. Watch between surveys. Use free imagery to track footprints weekly and flag unexpected changes.

  6. Referee quarterly. Once a quarter, laser-scan or ground-survey one pile on the same day as the drone flight, and record the difference as the programme's measured accuracy.

  7. Offer an independent view where needed. Where lenders or buyers need it, add satellite stereo on agreed dates, processed with the same boundaries and density.


When None of These Is Enough

Covered storage defeats every method above. Coal and fertiliser in domes or sheds need internal laser scanners or radar level sensors mounted inside, and the volume comes from those.

Piles that touch each other, with no visible toe between them, cannot be separated reliably by any surface method; agree with the yard team how the boundary is defined, and stick to it.

Material that changes density sharply with moisture, such as some ores and fines in the monsoon, needs density measured far more often than quarterly, or the tonnage will swing with the weather.

And where a transaction legally depends on a certified measurement, use the certified method the contract specifies, and treat the others as monitoring.


Back to the Three Numbers

The book stock, the drone survey and the satellite estimate for the coal yard were four percent apart. Most of that gap came from three choices that nobody had written down: the base surface each method assumed, the density each used to convert cubic metres to tonnes, and the boundary each drew around the piles. Agree those three, put the drone survey on fixed ground control, and the satellite becomes a useful independent check rather than a third opinion. The month-end meeting then argues about the one difference that matters: whether the weighbridge book has missed something real.


Write those choices down once, apply them to every method, and the three numbers become one stock figure the yard can plan against.

EurthTech is an engineering-first product design and embedded systems company with a development center in Hyderabad. To discuss a project, write to connect@eurthtech.com or see https://www.eurthtech.com/case-studies.

Comments


EurthTech delivers AI-powered embedded systems, IoT product engineering, and smart infrastructure solutions to transform cities, enterprises, and industries with innovation and precision.

Factory:

Plot No: 41,
ALEAP Industrial Estate, Suramapalli,
Vijayawada,

India - 521212.

  • Linkedin
  • Twitter
  • Youtube
  • Facebook
  • Instagram

 

© 2025 by Eurth Techtronics Pvt Ltd.

 

Development Center:

4th Floor, Krishna towers, 100 Feet Rd, Madhapur, Hyderabad, Telangana 500081

Menu

|

Accesibility Statement

bottom of page