Drone Stockpile Volumetrics: Accuracy and Verification

On This Page

• How the volume is calculated
• Why the buried base matters
• The commercial workflow and required inputs
• How to verify accuracy
• What the delivery package should contain
• Questions to settle before commissioning

How the volume is calculated

In photogrammetry, software matches features across overlapping photographs to reconstruct three-dimensional points. Those points support a surface model. For volumetrics, the operator also defines the pile's perimeter, often called its toe, and a base beneath it. PIX4D documents the relationship between image matching and reconstruction in its image acquisition guidance.

For a gridded calculation, each cell contributes its horizontal area multiplied by the difference between the top and base elevations. Summing those contributions gives the volume. PIX4Dmapper's calculation documentation describes this method using a digital surface model, or DSM. Other applications may use different surface representations, so record the actual method.

Volume ≈ sum of [cell area × (top elevation − base elevation)].

With area in square metres and height in metres, the result is cubic metres. Keep the sign convention explicit: material above the base is commonly reported as cut, while areas below it contribute fill. A net result can conceal those separate contributions. For a stockpile, unexpected fill is a reason to inspect the boundary and base before accepting the total.

Volume is also distinct from mass. Mass = volume × bulk density, using consistent units. Propeller's stockpile documentation treats material properties as additional inputs to the measurement. If the required output is tonnes, ask who supplies the density, how it was established, and which material condition it represents. A drone image does not supply that missing input.

Why the buried base matters

An overhead image sees exposed material. It does not reveal the ground underneath a solid pile. Software must use a surveyed base, a known elevation, or an inferred surface. These choices can produce different totals from exactly the same top surface.

The following table translates PIX4D's base-surface guidance into questions for the operator. The checks are editorial recommendations, not guaranteed accuracy classes.

A visible toe constrains the edge. It does not prove there is no depression or raised ground under the centre. Where that difference matters, arrange an empty-pad survey before loading or when the pad is next cleared. Preserve its version so future measurements use an identifiable reference.

A small height error across a large footprint

Consider an illustrative pile covering 2,000 m², with a reference volume of 8,000 m³. Suppose the adopted base is uniformly 0.10 m too low, while the top surface and footprint are otherwise correct.

Volume overstatement = footprint area × base-height error = 2,000 m² × 0.10 m = 200 m³.

Relative overstatement = 200 m³ ÷ 8,000 m³ × 100 = 2.5%.

These are hypothetical inputs, calculated September 7, 2026, with no rounding needed. They illustrate the surface-difference formula; they are not a measured drone result or an expected error rate. A nonuniform base error requires summing the differences across the footprint. The practical consequence is straightforward: improving the photographs cannot correct a wrongly specified buried surface.

The commercial workflow and required inputs

Begin with the output the site needs: one pile's current volume, a material inventory, or change between dates. Agree the reporting units, pile identifiers, measurement time, and how discrepancies will be handled. If loaders are adding or removing material during collection, the customer and operator need an agreed pause or a documented accounting boundary for those movements.

Establish the survey reference. Record the horizontal coordinate system, height reference, units, and base-surface source. Ground control points constrain the reconstruction; checkpoint coordinates are reserved for assessing its accuracy. PIX4D explains these distinct roles in its tie-point documentation. If aircraft positions use real-time kinematic or post-processed kinematic corrections, abbreviated RTK and PPK, retain the positioning records and still plan how the delivered surface will be checked.

Capture the surfaces that matter. Plan views of the crest, slopes, toe, and nearby stable ground. Image detail and overlap must be suitable for the material and geometry; an overlap setting alone is not proof of complete coverage. PIX4D's acquisition guidance highlights the difficulty of matching uniform sand and water surfaces. Inspect actual images for missing coverage, blur, deep shadows, and moving machinery, and supplement the collection where necessary.

Process and inspect the model. Retain the camera and processing settings, review alignment, and inspect the resulting surface for gaps, spikes, and objects that should not count as material. The USGS calibration report explains why flight geometry, camera calibration, control distribution, and quality assessment interact. A visually convincing orthophoto, the corrected overhead image, is not sufficient evidence of vertical accuracy.

Define and calculate each pile. Review the toe polygon and chosen base together. Preserve edits, exclusions, and the software version. For repeat work, reconcile pile splits, merged piles, and changed boundaries before comparing totals. An unexplained identifier change can make the inventory confusing even when each individual calculation is correct.

Keep the workflow proportional to the job. A routine operational estimate and a quantity used to settle a disputed delivery may require different independent verification. Agree that distinction before the flight so the necessary reference evidence can be collected.

How to verify accuracy

Separate image resolution, positional accuracy, volume agreement, and repeatability. They answer different questions.

Ground sampling distance, or GSD, describes the ground represented by an image pixel. It is a detail measure, not a measured volume error. PIX4Dmapper's volume-error estimate uses assumptions tied to GSD and explicitly excludes the effect of having few reconstructed points on an object. Treat a software error estimate according to its stated model, rather than as independent confirmation of the entire survey.

Check position without overstating what it proves

Request checkpoint locations, surveyed coordinates, coordinate differences, and a summary such as root mean square error, or RMSE. RMSE summarizes the size of discrepancies without positive and negative errors simply cancelling. Ask for horizontal and vertical results separately. Checkpoint coordinates used to adjust the model are no longer a withheld test of that adjustment. Their own survey quality and spatial coverage also limit what the check can establish.

Review where the checks are. Agreement on accessible ground does not directly test an unseen floor or every steep face. USGS distinguishes the quality supplied by control from the ability of check points to validate data; neither removes the need to inspect the product itself.

Absolute location and internal geometry also differ. DroneDeploy's accuracy guidance distinguishes relative measurements within a model from its position in the world. If the top and base share the same vertical shift, that shift cancels in their height difference. If today's top is compared with an independently surveyed historical base, a mismatch between their height references can change the calculated volume. Confirm their alignment on stable surfaces.

Compare volumes on equivalent terms

For a consequential job, commission a suitable independent reference measurement, such as a terrestrial laser scan or a survey designed to capture the relevant surface geometry. Match the measurement time, toe, base, exclusions, and units. Investigate disagreements in those inputs before assigning the difference to the drone.

Absolute percentage difference = |drone volume − reference volume| ÷ reference volume × 100.

Use that expression for a positive reference volume. Near zero, report the difference in cubic units because a percentage becomes unstable or undefined. Agreement with a reference is not automatically error against true volume: the reference has its own sampling and measurement limitations. Two methods that share the same incorrect base can agree closely.

Davis and Guy's 2023 aggregate-stockpile study compared UAV photogrammetry, terrestrial laser scanning, and total-station measurements across different aggregates. The authors reported differences in suitability by material and observed under-reporting with the total-station approach. That finding supports checking the reference method's fitness for the pile; it does not establish a universal percentage for drone surveys.

Finally, repeatability means obtaining similar results when the pile has not changed. Repeat flights can reveal inconsistency but can also reproduce the same bias. For change monitoring, ask the supplier how small a change the complete workflow can distinguish from survey variation. Do not assume the uncertainty of a difference equals that of one survey; shared errors and independent errors behave differently. The broader distinction between imagery and a supported measurement is covered in what drone inspection evidence can establish.

What the delivery package should contain

Use the handover to make the number reproducible. The following is a recommended commercial scope based on the documented calculation, control, and export workflows above; it is not a claim that every provider includes these items.

Request a sample export before commissioning. Confirm that the receiving GIS, CAD, or inventory system can read the proposed formats and coordinate references. A PDF is useful for review, but it cannot substitute for calculation inputs when the customer must reproduce or challenge the quantity.

Questions to settle before commissioning

Ask the supplier to answer these questions in the scope of work:

• What exactly does the quoted accuracy describe, and against which reference will it be assessed?
• How will the buried base be established, including areas hidden by walls or adjacent piles?
• Which observations will independently check the surface, and what remains unobserved?
• How will ongoing material movement, incomplete coverage, or failed checks affect delivery and recollection?
• Which files and input versions will we receive, and can we reproduce the calculation after the service ends?

For an isolated pile on a documented pad, commission the capture and checks needed to substantiate that defined volume. For an irregular or unknown buried base, resolve the base evidence first or explicitly limit the use of the estimate. For repeat inventory, prioritize consistent references and explainable changes. Those requirements make an accuracy claim useful to the buyer.

Source notes

• PIX4D: Selecting the Image Acquisition Plan Type. Technical guidance on image matching, coverage, and difficult surfaces.
• PIX4D: How PIX4Dmapper calculates the Volume. Primary documentation for the gridded surface-difference calculation and cut/fill signs.
• PIX4D: Which Base Surface is recommended for the Volume Calculation?. Software-specific guidance for visible and obscured pile boundaries.
• PIX4D: Error estimation in volume calculation. Assumptions and limitations of its GSD-based error estimate.
• PIX4D: Tie points in photogrammetry projects. Technical definitions of control points and checkpoints.
• USGS: Guidelines for Calibration of Uncrewed Aircraft Systems Imagery. Government technical report, 2023, on calibration, collection geometry, and quality assessment.
• DroneDeploy: How Accurate is My Map?. Provider documentation distinguishing relative and absolute accuracy.
• Davis and Guy: An Assessment of Point Cloud Data Acquisition Techniques for Aggregate Stockpiles and Volumetric Surveys. Original conference research, 2023; material-specific method comparison.
• Propeller: Measuring Stockpiles, The Basics. Provider documentation on outlines, exports, comparisons, and material-property inputs.

Last checked: September 7, 2026.