🎓 Lesson 9
D5
Geological Plausibility Scoring Metrics
Geological plausibility scoring metrics are simple scores that tell us how likely a computer-generated orebody model matches real geology — like giving a 'reality check' to AI predictions.
🎯 Learning Objectives
- ✓ Calculate a structural plausibility score using dip deviation and regional trend constraints
- ✓ Analyze grade-continuity plausibility by applying variogram-based spatial coherence thresholds
- ✓ Design a composite plausibility index combining lithological fidelity, contact sharpness, and alteration zoning compliance
- ✓ Explain how plausibility scores reduce conditional bias in AI-driven grade control decisions
- ✓ Apply industry-defined plausibility thresholds to accept/reject drill-hole interpolation patches
📖 Why This Matters
In AI-powered orebody modeling, algorithms can generate mathematically smooth but geologically absurd shapes — like vertical ore lenses dipping 85° in flat-lying sedimentary basins, or high-grade zones appearing abruptly across impermeable shale barriers. Without geological plausibility checks, these artifacts propagate into misclassified reserves, unsafe stope designs, and costly dilution. Plausibility scoring is your first line of defense: it quantifies 'geological common sense' so AI supports — rather than replaces — expert judgment.
📘 Core Principles
Plausibility scoring rests on three pillars: (1) Structural fidelity — assessing whether modeled contacts, folds, or faults respect regional deformation history and kinematic constraints; (2) Litho-geochemical coherence — verifying that grade, alteration, and lithology co-vary within empirically observed bounds (e.g., potassic alteration rarely occurs >50 m from porphyry centers); and (3) Spatial statistical reasonableness — ensuring grade continuity aligns with experimental variograms derived from validated drill data. Scores are typically normalized to [0,1] via sigmoidal or percentile-based scaling against reference populations (e.g., historical drill intercepts, expert-reviewed sections). Crucially, plausibility is not accuracy — a model may be plausible yet still inaccurate if input data is biased; conversely, a low score always signals need for expert review.
📐 Structural Plausibility Score (SPS)
The Structural Plausibility Score evaluates angular deviation between modeled orebody dip and the dominant regional structural trend, penalizing outliers beyond geologically justifiable thresholds. It uses a bounded inverse-tangent scaling to emphasize moderate deviations while tolerating minor noise.
Structural Plausibility Score (SPS)
SPS = 1 / (1 + exp(k × (|θ_model − θ_regional| / θ_max − c)))Measures angular consistency of modeled orebody dip with regional structural trend, scaled to [0,1].
Variables:
| Symbol | Name | Unit | Description |
|---|---|---|---|
| θ_model | Modeled dip angle | degrees (°) | Dip angle of the AI-modeled orebody surface at the evaluation point |
| θ_regional | Regional structural dip | degrees (°) | Mean dip of major controlling structures (e.g., host strata, shear zones) in the domain |
| θ_max | Maximum allowable angular deviation | degrees (°) | Geologically justified tolerance (e.g., ±15° for sedimentary-hosted systems; ±8° for rigid-block tectonics) |
| k | Steepness parameter | dimensionless | Controls sensitivity; default = 2.5 for balanced discrimination |
| c | Centering offset | dimensionless | Threshold crossover point; default = 0.7 (i.e., SPS = 0.5 when deviation = 70% of θ_max) |
Typical Ranges:
Sedimentary exhalative (SEDEX): ±12° – ±18°
Porphyry Cu-Au: ±6° – ±10°
Orogenic gold: ±5° – ±15°
💡 Worked Example
Problem: Given: modeled orebody dip = 42°, regional structural trend dip = 28° ± 9° (1σ), rock type = porphyry copper deposit (typical allowable deviation = ±15°), σ = 9°.
1.
Step 1: Compute absolute angular deviation: |42° − 28°| = 14°.
2.
Step 2: Normalize deviation relative to allowable threshold: x = 14° / 15° = 0.933.
3.
Step 3: Apply sigmoidal scaling: SPS = 1 / (1 + exp(2.5 × (x − 0.7))) = 1 / (1 + exp(2.5 × (0.933 − 0.7))) = 1 / (1 + exp(0.583)) ≈ 1 / (1 + 1.791) ≈ 0.359.
Answer:
The result is SPS = 0.36, which falls below the recommended minimum threshold of 0.65 — triggering expert review of the modeled geometry.
🏗️ Real-World Application
At Newcrest’s Telfer Mine (Western Australia), an AI grade-control model generated a narrow, steeply dipping (>65°) gold zone within a known shallow-dipping, bedding-controlled mineral system. The structural plausibility score was 0.21 (well below the site-minimum 0.6), prompting geologists to re-examine drill orientation errors and confirm that the anomaly arose from uncorrected down-hole survey drift. Correcting the survey data and re-running the model raised SPS to 0.83 — restoring geological credibility and preventing a 12% overestimation of high-grade tonnage in the next blast round.
🔧 Interactive Calculator
🔧 Open AI-Powered Orebody Delineation & Grade Control Calculator📋 Case Connection
📋 Copper Mine Block Model Refinement Using Neural Kriging
Traditional kriging over-smoothed high-grade chalcocite zones, causing 8.2% reserve underestimation
📋 Gold Mine Real-Time Grade Control at Development Drift Face
Manual chip sampling caused 24–48 hr delay in stope boundary decisions, leading to 14% dilution
📋 Iron Ore Mine Sensor Fusion for Banded Iron Formation (BIF) Delineation
Conventional geophysics failed to resolve thin hematite bands (<2m) within jaspilite, causing 22% grade variance in ROM...
📋 Limestone Mine Digital Twin for Karst-Related Grade Uncertainty
Solution cavities caused unpredictable grade drops (CaCO₃ purity <85%) in otherwise uniform deposits, resulting in 11% p...