🎓 Lesson 14
D5
Quantifying Geochemical Prediction Uncertainty with Monte Carlo Methods
Monte Carlo methods use repeated random sampling to estimate how uncertain predictions are when we don’t know exact values for geochemical inputs like mineral content or pH.
🎯 Learning Objectives
- ✓ Calculate 95% prediction intervals for Net Acid Generation (NAG) pH using Monte Carlo simulation outputs
- ✓ Design a Monte Carlo workflow in Python or GoldSim for a simple static geochemical model with ≥3 uncertain inputs
- ✓ Analyze parameter sensitivity using standardized regression coefficients or Sobol indices derived from simulation results
- ✓ Explain how input distribution assumptions (e.g., lognormal vs. uniform) impact risk classification outcomes
- ✓ Apply regulatory thresholds (e.g., BC MESC or EPA ARD guidelines) to interpret simulated exceedance probabilities
📖 Why This Matters
In mine waste characterization, a single 'best guess' pH or NAG value can mislead decisions—leading to under-designed water treatment, unexpected ARD, or costly over-engineering. Real-world samples show high variability: pyrite content may range from 0.1% to 8% across a waste dump, and carbonate buffering varies by orders of magnitude. Monte Carlo methods transform this variability into actionable risk insight—telling engineers not just *what* might happen, but *how likely* it is. Regulators (e.g., BC EMA, US EPA) now expect quantitative uncertainty analysis in closure planning—and courts have cited inadequate uncertainty treatment in liability cases.
📘 Core Principles
Uncertainty in geochemical prediction arises from three sources: measurement error (e.g., ±0.3% on sulfide S assay), spatial heterogeneity (e.g., nugget effect in drill core assays), and model structural uncertainty (e.g., simplified kinetic rate laws). Monte Carlo addresses parametric uncertainty by treating each input as a probability distribution—not a fixed number—then running hundreds to thousands of model evaluations. Each run produces one output value; the ensemble forms an empirical output distribution. Key theoretical foundations include the Law of Large Numbers (ensuring convergence of statistics) and Central Limit Theorem (justifying normality of means for large N). For geochemical models, Latin Hypercube Sampling (LHS) is preferred over simple random sampling to ensure efficient coverage of multidimensional input space with fewer runs (typically 500–2000). Sensitivity analysis then identifies which inputs drive most output variance—guiding where to invest in better data collection.
📐 Monte Carlo Prediction Interval
The 95% prediction interval quantifies the central 95% of simulated output values—directly conveying the plausible range of outcomes given input uncertainty. It is computed non-parametrically from the empirical distribution of Monte Carlo results.
Empirical Percentile Interval
PI_{1−α} = [x_{(⌊α/2·N⌋)}, x_{(⌊(1−α/2)·N⌋)}]Non-parametric confidence/prediction interval capturing central (1−α)% of Monte Carlo output values.
Variables:
| Symbol | Name | Unit | Description |
|---|---|---|---|
| PI_{1−α} | Prediction interval | unitless or pH | Range containing central (1−α)% of simulated outputs |
| x_{(i)} | i-th ordered output value | pH, kg/t, etc. | Sorted output from Monte Carlo simulation |
| N | Number of simulations | dimensionless | Total runs used (typically 500–5000) |
| α | Significance level | dimensionless | Commonly 0.05 for 95% interval |
Typical Ranges:
Regulatory reporting (BC MESC): 90% or 95% PI required for NAG pH
Kinetic modeling runtime constraints: N = 1000–2000 for balance of accuracy & compute time
💡 Worked Example
Problem: A Monte Carlo simulation (N = 1000 runs) of NAG pH for a waste rock unit yields sorted pH outputs. The 2.5th percentile = 3.12, and the 97.5th percentile = 4.89.
1.
Step 1: Sort all 1000 simulated pH values in ascending order.
2.
Step 2: Identify the index positions: 0.025 × 1000 = 25th value (3.12) and 0.975 × 1000 = 975th value (4.89).
3.
Step 3: Report interval as [3.12, 4.89] — meaning there is 95% probability the true NAG pH falls within this range, assuming input distributions and model structure are valid.
Answer:
The 95% prediction interval is [3.12, 4.89], which exceeds the BC MESC threshold of pH < 4.5 for ARD concern — indicating >97.5% probability of acidic drainage generation.
🏗️ Real-World Application
At the Mount Polley tailings facility (BC, Canada), post-2014 failure investigations used Monte Carlo simulation (GoldSim) to quantify uncertainty in long-term pore water pH. Inputs included lognormal distributions for total sulfide (mean = 0.8%, σ = 0.4), carbonate content (uniform: 0.5–3.2%), and weathering rate constants (triangular: 0.001–0.015 d⁻¹). With 1500 LHS runs, the model showed a 68% probability of pH < 4.0 over 100 years—triggering redesign of the cover system to include alkaline amendment. This analysis directly informed the 2021 Updated Closure Plan accepted by BC EMA.
🔧 Interactive Calculator
🔧 Open Mine Waste Characterization & Geochemical Modeling Calculator📋 Case Connection
📋 Copper Mine Waste Rock Stockpile ARD Mitigation at Escondida Extension
High-pyrite waste rock (>3.2% S) stockpiled without cover; predicted ARD onset within 5 years
📋 Gold Tailings Geochemical Stabilization at Granny Smith Mine (WA)
Arsenic-rich tailings (up to 120 mg/kg As) exhibiting elevated As leaching under oxidizing conditions
📋 Iron Ore Mine Waste Rock Long-Term Stability at Brockman 4 (Pilbara)
Massive hematite-goethite waste rock (low sulfide but high Mn/Al) showing delayed acidity and Al leaching post-construct...
📋 Coal Mine Spoil Geochemical Capping at Hunter Valley Reclamation Project
Spoil with pyritic shale interbeds generating ARD despite initial alkaline overburden; inconsistent capping led to local...