Chapter 03
How exploration is done
The physical workflow that produces every number you will ever model. Read this as a data-generating process: each step decides what gets measured, what does not, and what gets quietly lost.
Stage 0 · Get the ground
You cannot explore land you do not control. Mineral rights are usually separate from surface rights, and are acquired as claims (North America) or tenements / exploration licences (Australia, Africa). Most jurisdictions have moved to online map staking, so claims are now a dataset — polygons with owners and expiry dates.
Claims must be kept alive by spending money or doing work each year (assessment work). This produces two useful side effects. Claim boundaries are open data in most jurisdictions and tell you where professionals currently believe value lies. And expiring ground is a real signal — somebody just decided it was not worth the annual fee.
Claim polygons correlate strongly with prospectivity, because they encode expert opinion plus historical discovery. Using them as a feature will inflate your metrics and teach the model nothing about geology. It is the same disease as distance-to-road, wearing a lab coat. See Chapter 9.
Stage 1 · Desktop study
Before anyone flies anywhere, someone spends weeks assembling what is already known: government survey geophysics and geochemistry, historical assessment reports filed by previous operators, old drill logs, academic theses, satellite imagery. Geological surveys — the USGS, Geoscience Australia, provincial and state surveys — release enormous quantities of this for free.
This stage is unglamorous, slow, and where a genuinely large fraction of discoveries actually begin. It is also the stage most obviously amenable to automation, because the bottleneck is reading and reconciling thousands of inconsistent documents rather than any geological subtlety.
Stage 2 · Reconnaissance
Now you cover large areas cheaply, accepting low resolution.
- Airborne geophysics. Fly magnetics, radiometrics and sometimes electromagnetics on a regular flight-line grid. Cheap per square kilometre and the standard first look at covered ground.
- Stream sediment sampling. Sample the sediment in drainages; each sample integrates the catchment upstream of it. A beautifully efficient sampling design — a few hundred samples can screen thousands of square kilometres — and one whose spatial support is a catchment polygon, not a point. Treating stream sediment samples as point observations is a common and serious error.
- Regional mapping. Geologists walk the ground and record rock types, structures and alteration.
Stage 3 · Target generation
Anomalies from reconnaissance get followed up at higher resolution: soil sampling on a regular grid, ground geophysics (IP, EM, gravity) along lines across the target, detailed mapping, rock chip sampling of outcrop, sometimes trenching or auger drilling through shallow cover.
The output is a ranked list of drill targets, each with a proposed collar location, azimuth, dip and depth, and a stated hypothesis about what it should intersect. That hypothesis is the object that Chapter 10 teaches you to test efficiently — and, when necessary, to falsify early.
Stage 4 · Drilling
The moment of truth, and the point where cost changes character. Two methods dominate.
| RC (reverse circulation) | Diamond core | |
|---|---|---|
| Returns | Rock chips, blown up the rod string | A continuous cylinder of intact rock |
| Speed | Fast — can exceed 100 m/day | Slower |
| Cost per metre | Lower | Higher, often several times |
| Gives you | Grade, rough lithology | Grade plus structure, texture, contacts, orientation, geotechnical data |
| Depth limit | Practically a few hundred metres | Kilometres |
| Use it for | Cheap grade information over known geology | Anything where you need to understand the rock, and all resource-definition work |
A drill programme is a logistical operation: permits, access tracks, water supply, a camp, a contractor with a rig and a crew on shifts. Mobilisation alone can cost as much as several holes. This is why drilling comes in campaigns — and why, as Chapter 6 shows, campaign structure is stamped into your data as a spatio-temporal batch effect.
What happens to the core
core out of the hole
→ oriented and marked, laid into trays in order
→ photographed (wet and dry)
→ GEOLOGICAL LOGGING lithology, alteration, mineralisation,
structure, veining, recovery, RQD
→ sometimes: core scanning (hyperspectral, XRF, density)
→ sampling intervals marked, honouring geological contacts
→ CUT IN HALF half to the lab, half retained forever
→ bagged with a unique sample number
→ QA/QC control samples inserted blind into the sequence
→ shipped to the assay laboratory
→ results returned as a certificate, loaded to the database- Logging is human and subjective. Two geologists log the same core differently, and a single geologist logs differently in week one and week ten. Lithology codes are noisy labels with a rater effect — and the raters change between campaigns.
- Sampling intervals are chosen, not fixed. Geologists sample to geological contacts, so interval length is itself informative and correlated with mineralisation. Your "missing at random" assumption is already false.
- Half the core still exists. This is a gift: unlike almost any other domain, you can go back and re-measure the original physical specimen years later. Re-assay programmes and core re-scanning are real options for resolving a data dispute.
Stage 5 · Interpret, and decide
Results come back weeks later. The geologist updates the model, and the company faces the only question that matters: drill again, and where — or stop?
In practice this decision is made in a room, from a combination of the new intercepts, the geological story, the treasury balance and the market's appetite. It is rarely made with an explicit expected-value calculation, and almost never with a formal statement of what would change the answer. That gap is the opportunity that Chapter 10 is about.
Permitting, ESG and social licence
Not a footnote — routinely the binding constraint on whether a deposit ever becomes a mine, and increasingly on whether you can drill at all.
- Permits for ground disturbance, water use, drill pads and access, at several levels of government.
- Indigenous and community consultation. In Canada and Australia there are legal duties to consult; in most jurisdictions there is a practical requirement regardless. Agreements are negotiated, take time, and can restrict where and when you work.
- Environmental baseline studies — water, flora, fauna — often started years before they are needed, because you cannot retroactively collect a baseline.
- Social licence, the informal version of all of this: a project with local opposition can be stopped after permits are granted.
Access constraints appear in the data as holes in the coverage that have nothing to do with geology — a national park, a pastoral lease, a contested boundary, a lake. Model those explicitly as excluded ground rather than letting them masquerade as low-prospectivity, and never present a target the company is not legally able to drill.
The disclosure cycle
Listed companies must disclose material information promptly, which produces the rhythm of the industry: a drill result arrives, is verified, and is released as a news release with a table of intercepts. Larger milestones — a maiden resource, a feasibility study — require a full technical report under NI 43-101 or JORC, signed by a Qualified Person.
These documents are the industry's public corpus and a genuinely rich, under-exploited dataset: SEDAR+ (Canada) and the ASX announcements platform hold decades of technical reports, each containing drill tables, QA/QC discussion, resource statements and interpretation. They are also legally constrained documents, which is why the vocabulary in Chapter 4 is worth taking seriously.
What a season actually looks like
Field programmes are seasonal — snow, monsoon, heat, road access. A northern Canadian project might have a four-month window; the rest of the year is compilation, planning, financing and modelling.
This is worth knowing for a practical reason. There is a narrow window each year in which a model can influence the drill plan, and outside it your work will be filed and forgotten. Ask, early, when the programme is being planned and when the budget is committed. A good answer delivered two weeks after the rig is booked has a value of information of exactly zero — which is the Chapter 10 lesson arriving early and in person.
Sources for this chapter
- Marjoribanks, Geological Methods in Mineral Exploration and Mining (Springer, 2nd ed. 2010). The standard practical field manual — mapping, logging, sampling, drilling. Closest thing to "how the work is actually done."
- Abzalov, Applied Mining Geology (Springer, 2016). Drilling methods, sampling protocols, QA/QC design and resource-estimation practice, from a practitioner.
- Moon, Whateley & Evans, Introduction to Mineral Exploration (Blackwell, 2nd ed. 2006). Broader survey covering the whole workflow including economics and evaluation.
- USGS Mineral Resources Online Spatial Data and Geoscience Australia data & publications — the kind of public survey data a desktop study starts from.
- SEDAR+ (Canada) and the ASX announcements platform — every NI 43-101 and JORC technical report, free. The largest public corpus of real exploration data and interpretation that exists.
- CIM — NI 43-101 standards and definitions and the JORC Code. The actual regulatory texts; skim the definitions sections at least once.