Gains:
- Ability to construct geostatistical steps such as compositing, variogram and kriging with AI and question assumptions
- Ability to interpret block model, grade-tonnage curve and resource classification (measured/displayed/extracted) with AI support
- Ability to test prediction and uncertainty values produced by AI through independent calculation, cross-validation and qualified person (CP/QP) judgment
The most critical number for a mining company is how much and what grade of ore is underground. This number; It determines the investment decision, stock market value, bank financing and the life of the furnace. But no one can measure all the substance one by one; only sparse drilling points are sampled and intervals are estimated. The discipline that places this prediction on a scientific basis is called geostatistics. Geostatistics models how similar spatial (based on location in space) data are to each other and estimates unmeasured points with a margin of uncertainty. AI; It is a powerful aid in the code, scenario comparison and uncertainty analysis parts of this process. However, it is the decision of the competent person (Competent Person / CP in JORC, Qualified Person / QP in NI 43-101) which model is valid, which assumption is acceptable and how the resource will be classified.
Resource and reserve: critical distinction
Let's clarify the two terms first. Mineral Resource: Ore that can be extracted economically based on geological information; They are classified as inferred, indicated and measured according to the confidence level. Mineral Reserve (reserve): the part of the resource that can actually be extracted profitably through technical and economic work (mining method, facility efficiency, cost, metal price); They are classified as probable and proven. The resource is geological, the reserve is economic. AI can explain these definitions, but it cannot decide which class a number falls into; This is the judgment of the competent person looking at data density and continuity confidence.
Step by step: geostatistical workflow and the place of AI
- Compositing. Bringing assay intervals of different sizes to equal length (e.g. 1 m or 2 m). AI: writes code, establishes unit and weighted average logic; You control the outcome.
- Statistical review. Grade distribution, outliers, capping (top-cut: limiting extremely high grades). AI: generates histogram/summary statistics code; The geologist makes the capping decision.
- Variogram analysis. Modeling spatial continuity: how similarity decreases as points move further apart. Range, sill and nugget parameters. AI: explains the concept of variogram, gives a code skeleton; The geostatistician verifies model fit.
- Prediction (kriging/IDW/simulation). Assigning grades to each block of the block model. Kriging is a weighted estimation method based on variogram. AI: explains method differences, builds code; The expert chooses the parameters.
- Verification. Cross-validation, prediction-to-sample comparison, visual inspection, smoothing control. AI: generates validation graphs; The expert makes the comment.
- Classification and reporting. Resource/reserve classes, grade-tonnage curve, JORC/UMREK compatibility. AI: draft text and table; The signature belongs to the authorized person.
Tip: Don't make the AI say "you choose the variogram parameters". Range and nugget impact depend on the geology of the ore body; the wrong choice systematically inflates or shrinks the reserve. Use AI as code and scenario generator, leave the decision to the geostatistician.
Grade-tonnage curve and economic limit grade (cut-off)
Cut-off grade is the threshold that determines whether a block is considered ore or rust; metal price depends on cost and plant efficiency. As the limit grade changes, the tonnage and average grade of the reserve changes; This relationship is shown by the grade-tonnage curve. AI can write code that calculates this curve and quickly compare scenarios (different cut-off, different metal price). But the cost and price assumption on which the cut-off is determined is an economic decision and is made together by the technical team and management.
three mini cases
Case 1 — Capping negligence. In a gold project, several drillings yield exceptionally high grades of 120 g/t. The team krigs them without capping; The average grade is 2.4 g/t. AI is cross-validated; Estimates are actually inflated by very little drilling, with a few blocks being assigned extremely high. After appropriate capping (e.g. 30 g/t) the average decreases to 1.7 g/t. The difference directly changes the feasibility of the project. AI produced the calculation and chart; The capping level decision was made by the geostatistician.
Case 2 — Wrong range. If the variogram range in a copper porphyry is actually 150 m but the model is taken to be 400 m, the sparse drilling areas appear too "reliable" and the blocks that should be considered inferred will shift to the indicated class. This is detected in an audit and a significant portion of the resource is downgraded to a lower trust class. The AI had generated the variogram code, but the error was magnified because the range was not reviewed by an expert; The lesson is to ensure that model fit is verified by an expert.
Case 3 — Correct scenario comparison. In an iron project, the team wants to see grade-tonnage results for three different cut-offs (20%, 25%, 30% Fe). They write a Python script to the AI that does the mass/volume/density logic and grade-tonnage calculation, using their own block model data. Script tables 3 scenarios in seconds. The team validates the results by hand on a block and walks into the decision meeting with a clear comparison chart. AI has accelerated the iterative calculation here; The data and decision belonged to the team.
Copiable prompt templates
COMPOSITION CODE "Role: You are a geostatistics assistant. Write a function with Python + pandas that converts from-to-interval assay data into FIXED [1 m] composites. Rules: WEIGHTLY average the grade by interval length; skip and report missing/empty intervals; process separately based on hole ID. Write the unit assumption at the beginning with a comment line. Return how many intervals were skipped in the result."
VARIOGRAM CONCEPT + CODE "What is a variogram: explain the concepts of nugget, sill and range in simple language. Then calculate the experimental variogram in one direction and give a spherical model fitting rank code skeleton. Add a WARNING note: range and nugget must be expertly selected according to the geology of the ore; DO NOT SELECT these values."
GRADE-TONNAGE SCENARIO "The block model yield contains the following columns: [x,y,z, grade, blockvolume, density]. Write a script that calculates the total tonnage and average grade in each scenario for the following cut-off values [0.3; 0.4; 0.5]. Explicitly interpret the density and unit assumption. Return the result as a comparison table."
CROSS-VALIDATION COMMENT "Below are the kriging cross-validation results (actual vs prediction). Comment on: is there systematic bias, are the predictions oversmoothing, in what grade range is the deviation large? DO NOT make a firm decision; suggest which parameter the geostatistician should review. Data: [paste]."
Weak prompt / Strong prompt
WEAK PROMPT: "Calculate the reserve for this project."
STRONG PROMPT: "Role: You are the geostatistics assistant. I will give you my own block model and variogram parameters. Set up a Python flow with interpreted units and assumptions that produces kriging estimation + cross-validation + grade-tonnage curve. SELECT parameters; use mine. Make a note at the end of the report that the results should be classified and approved by a qualified person (CP/QP) according to JORC/UMREK."
Comparison table: method and AI role
step
Purpose
AI role
Expert judgment
Compositing
Equal support length
Code + control
Composite size
Capping
Outlier limit
Graph/statistics
Cap level
variogram
spatial continuity
Concept + code
Range/nugget selection
kriging
Block prediction
Code + stream
Parameter + method
Classification
Trust level
draft text
JORC/UMREK judgment
Common mistakes
- Skipping capping. A few bonanzas inflate the value average and reserve.
- Having the variogram parameters selected by AI. Range/nugget depends on geology; Wrong choice systematically disrupts the reserve.
- Confusing resources with reserves. The resource is geological, the reserve is economic; The classes are not interchangeable.
- Ignoring smoothing. Kriging may over-soften the tenor; It reduces high tenor and enlarges low tenor.
- Reporting the classification without competent person approval. JORC/UMREK is valid only with the signature of an authorized person.
Attention: The reserve number is an official declaration disclosed to the stock exchange, bank and investor. No intermediate result produced by AI can become an official reserve/resource declaration without the independent evaluation and signature of a competent person (CP/QP).
In summary
Geostatistics is the discipline of estimating the entire ore body with uncertainty from sparse drilling: compositing, capping, variogram, kriging, verification and classification. In this stream, AI is strong in coding, scenario comparison, grade-tonnage calculation and verification graphics. But variogram parameters, capping level and resource/reserve classification are expert decisions and the official declaration requires the signature of the competent person. The resource is geological, the reserve is economic; Never blur this distinction.
Application task
Produce tonnage and average grade tables for three different cut-offs using the "Grade-tonnage scenario" template with a sample (or your own) composited grade data. Then have the AI interpret a kriging result with the “Cross-validation interpretation” template and list which parameters need expert review. Finally, have the AI explain the difference between resource and reserve and classification confidence levels in your own words and check its accuracy against the official code definition.
checklist
- [ ] I determined the composite size and unit assumptions, I did not leave it to AI.
- [ ] The geologist/geostatistician made the capping decision for extreme values.
- [ ] Variogram range and nugget value expert reviewed.
- [ ] I tested the Kriging result with cross-validation and visual inspection.
- [ ] I used resource and reserve classes correctly and separately.
- [ ] I noted in the report that competent person (CP/QP) approval is required for official declaration.