Sampling Design and Representativeness

Sampling design determines which material is observed, how it is selected and what population the resulting measurements can describe. Representativeness is a relationship between a sample, a defined target and a property of interest; it is not a label earned simply by collecting a large bag or obtaining repeatable assays. When the intended result describes a larger target, uncertainty arising from sampling belongs in the measurement assessment.^[1]

A useful design begins with the engineering decision. It identifies the material and property that matter, the spatial or temporal scale of the conclusion, and the consequences of an incorrect estimate. Only then can sample locations, collection methods, masses and quality checks be justified.

Defining the Target and the Property

A target can be a geological interval, an excavation volume, a stockpile lot or material passing a transfer point during an identified period. Its boundaries determine what belongs in the estimate. The property also needs definition: an elemental mass fraction on a dry basis, for example, differs from an extractable fraction measured by a specified digestion. Eurachem's sampling guidance distinguishes the concentration in a sampling target from the concentration in the sample delivered to a laboratory.^[1]

The design should distinguish the physical sample from the engineering volume represented by interpretation. A core assay directly characterizes submitted material from a narrow intersection. Extending its meaning into surrounding ground requires geological and spatial assumptions; it is not an automatic consequence of competent laboratory work. CIM treats data quality, spatial distribution and adequacy for estimation as separate considerations.^[2]

A sample can also be appropriate for one property but inadequate for another. A selection intended to describe average chemical composition may omit the range of textures or material types needed for processing tests. The resulting test cannot be generalized merely because the composite grade resembles the planned feed. The target population and selection rationale need to match each intended use.

Matching Coverage to Geological Variability

Coverage should address the geological domains and directions of variation relevant to the question. Closely spaced observations in an accessible area do not compensate for an unobserved zone that could change the interpretation. Nor does a large number of samples establish continuity when their positions, recoveries or support are poorly known.^[2]

Sample support describes the size, shape and orientation over which a value is measured. Changing interval length or combining intervals changes the scale of averaging. Small high-grade features may be visible in short samples but obscured in longer composites. CIM's guidance therefore relates compositing to geology, the data and the intended mining context rather than prescribing a universal interval length.^[2]

Geological mapping can guide targeted investigation, but a deliberately selected mineralized specimen is not an unbiased estimate of the surrounding excavation. Its legitimate role may be mineral identification or hypothesis testing. Recording that purpose prevents a useful diagnostic sample from later being mistaken for a production-grade observation. Planned samples, selective specimens and missing intervals should remain distinguishable in the database.

Material Heterogeneity and Selection

At particle scale, constituents may be concentrated in particular grain sizes, mineral species or densities. Segregation during handling can then separate those constituents spatially. If a collection device misses fines or preferentially admits small fragments, its output may differ systematically from the target. Minnitt's minerals-industry review describes these selection and extraction mechanisms and emphasizes that a nominal sampling method does not guarantee correct implementation.^[3]

For particulate lots, equal opportunity of particle inclusion is an important design principle, with weighting and increment collection appropriate to the lot.^[3] This should not be confused with an assertion that every geological location is equally accessible or that all geological inference must use a uniform grid. Targeted geological investigation and estimation of a lot average require different sampling justifications.

Increasing mass can help address some particle-scale variability, but it cannot compensate for selecting only one conveniently accessible part of a heterogeneous lot. Increasing spatial coverage addresses a different problem. The design must therefore distinguish the number and distribution of primary increments from the mass retained at each subsequent preparation stage.^[3]

Collection, Compositing and Weighting

Collection rules should define the intended increment and how it is recovered, including treatment of boundaries, fines, moisture and interruptions. A sampled fraction must not become progressively more selective as material moves from the field to a laboratory portion. Eurachem represents preparation and subsampling as parts of the complete measurement process, not as neutral transfers outside its uncertainty.^[1]

Physical compositing mixes material before analysis; numerical compositing combines measured values afterward. Both need weights consistent with the quantity being estimated. Equal masses from different production periods create an equal-mass mixture of those increments, not automatically the average composition of all tonnes produced during the periods.

For illustration, suppose two fully represented dry lots contain 10 and 30 tonnes at 1 and 3 g/t respectively. Their combined grade is 2.5 g/t, because the total contained quantity is divided by 40 tonnes. The unweighted mean, 2 g/t, answers a different question. This hypothetical calculation assumes consistent dry-mass grades and no unaccounted material transfer; it is not a prescription for combining spatial samples. With estimated grades or masses, the arithmetic does not remove uncertainty in those inputs.

Length weighting can be useful for defined geological composites. Equivalence to a mass-weighted result depends on the cross-section and density of the material represented; a recovered sample's mass is not automatically that target mass. Weighting by recovered length or mass cannot establish the unknown grade of selectively lost material. Combining intervals must also preserve original measurements and respect the intended geological domains.^[2][1]

Testing the Design

Initial investigations can compare collection methods, examine material recovery and test how variability changes with retained mass or preparation. Their results should cover relevant material types rather than only the easiest or most uniform samples. A design demonstrated on one matrix or grain-size distribution may need reassessment when those conditions change.^[4]

Duplicate sampling can quantify variability under the conditions repeated. The Eurachem guide explains why a single protocol repeated by one sampler does not necessarily reveal bias shared by both samples. Repeated laboratory measurements provide an even narrower test if they start from the same prepared material. Agreement is evidence about the repeated process, not proof of complete representativeness.^[1]

Independent observations and reconciliation can provide further checks, provided their own supports, periods and uncertainties are understood. A difference between a geological estimate and production is a prompt to examine collection, modelling and material accounting; it does not identify the sampling error by itself.

Implementation and Revision

The implemented design needs records of locations, dates, intervals or lot boundaries, collection conditions, identifiers, retained fractions and deviations. Secure handling and documented custody preserve the link between the designed sample and the material actually analysed. CIM's exploration guidelines connect these responsibilities to trained personnel and verification throughout a campaign.^[4]

Revisions should be triggered by evidence: a new mineralization style, changing recovery, an inaccessible target area, a different handling system or a new decision requirement. The revised design should identify which earlier data remain comparable. A credible outcome is not a declaration that sampling error has disappeared, but a documented explanation of what the samples represent, what the checks demonstrate and where the remaining uncertainty affects engineering use.