Underground Gold Mining Techniques and Comparison

The nugget effect deserves a paragraph to itself because everything downstream in method selection depends on it. On a variogram, the nugget is the y-intercept, the variance that exists at zero lag distance. In porphyry copper deposits the nugget-to-sill ratio is commonly 0.1 to 0.3. In orogenic gold deposits, Cornwell (2005, Ph.D. thesis, University of Queensland) documented ratios of 0.4 to 0.8 across eleven Australian gold deposits. Vann and Guibal (1998, in the proceedings of the Symposium on Beyond Ordinary Kriging) pointed out that at nugget ratios above 0.5, Kriging is essentially computing a moving average that barely differs from inverse-distance weighting. The smooth contour maps in feasibility reports do not convey this. They look precise. The underlying data does not support that precision at the scale of a single stope.

This has a direct consequence for method selection that the rest of this article keeps coming back to. Any method that requires committing to stope boundaries months before production, based on the block model, is carrying the full risk of model error. Any method that allows adjustment in real time during extraction is partially hedged against model error. That is not a minor distinction. On orebodies with nugget ratios above 0.6, the block model can misclassify entire stopes from ore to waste or waste to ore.

Sericite and chlorite alteration around gold veins make the rock mechanics picture worse. Both minerals lose strength when wet. Friction angle drops of 30% to 50% after saturation are documented in Marinos and Hoek's 2001 GSI tables for weak rock. The issue is that exploration-stage testing uses dry or natural-moisture samples, and the RMR or Q-value assigned to the rock mass reflects dry conditions. After a few years of mining, groundwater migrates into the disturbed zone, altered rock stays saturated, and the actual rock mass quality can be a full class below the design assumption. Multiple mines in the Chilean and Peruvian Andes have experienced this. The fix, designing to wet-state parameters from the start, is known. It adds cost to the feasibility study because saturated testing takes longer. Many projects skip it.

Stress measurement is another area where feasibility-stage data is thin. Three to five measurement points, typically in competent rock near the decline, extrapolated over the whole orebody. In structurally complex gold deposits the stress tensor can rotate across a single fault. Stoping sequences designed on the assumption of a uniform stress field have had to be torn up and redesigned after production revealed localized high-stress conditions near structures. This happened in Ontario's deep gold belt and the result was write-offs of already-developed headings, disrupted backfill schedules, and production shortfalls.

Cut and fill gets far more space in this article than the other methods because the economics of grade control in gold mining make it disproportionately important. Not because it is the best method in general. It is slow, expensive per tonne, labor-intensive, and bottlenecked by its backfill system. In base metal mining nobody uses it unless ground conditions force the choice. In gold mining the grade control it provides has a measurable dollar value that can, in high-grade ground, exceed the entire cost premium over faster methods.

The mechanism is straightforward. Mining proceeds upward layer by layer. Each layer is backfilled before the next is taken. The miner sees fresh orebody at every lift and can shift the heading left, right, wider, or narrower depending on what the mineralization looks like. This matters when the grade coefficient of variation in the orebody exceeds 150%, which at orogenic gold deposits it routinely does. At Red Lake, Pakalnis and Hughes (2011, CIM Journal) reported stope-scale grade CVs above 250% in parts of the high-grade zone.

Reconciliation data from cut-and-fill gold mines consistently shows mill feed grade running above reserve model grade. This is positive mining bias. The block model, constrained by the nugget effect, smooths highs and lows into a blended estimate. The miner in the heading sees where quartz is thick, where sulfide concentration changes, where the vein splits, and preferentially chases rich ground. The reserve model does not account for this because there is no defensible way to forecast it during feasibility. It shows up after commissioning as a persistent positive variance in the mill-to-model reconciliation. On high-grade gold deposits it can contribute multiple millions of dollars annually in revenue that the feasibility study never promised.

The cost side is real. Cycle time includes drilling, blasting, mucking, filling, and curing. Cemented paste fill cures in 7 to 28 days depending on cement content and binder type. The stope produces nothing during cure time. Throughput per stope is low. At Australian underground mining labor rates (A$120,000 to A$150,000 per operator per year, fully loaded), the cost per tonne underground can run A$80 to A$130 in cut-and-fill stopes versus A$25 to A$50 in open stoping. Whether the grade control premium justifies that cost difference depends on the orebody. Above 8 g/t it almost always does. Below 3 g/t it almost never does. Between 3 and 8 the answer is specific to the deposit and sensitive to gold price.

The other two methods that account for the majority of underground gold production both involve some form of bulk extraction followed by (in the case of open stoping) or without (in the case of sublevel caving) backfill. Both accept more dilution than cut and fill. The question is how much dilution, and what it costs in gold terms.

Sublevel caving: 15% to 25% dilution is the standard textbook range and it holds up in practice. Laubscher and Jakubec (2001, Underground Mining Methods, SME) give similar ranges. In gold, the dilution penalty is nonlinear. A tonne of copper ore diluted from 1.2% Cu to 1.0% Cu still reports to the concentrator and still recovers metal, at a lower recovery and lower concentrate grade, but the revenue decline is roughly proportional to the grade decline. A tonne of gold ore diluted from 6 g/t to 5 g/t may still be above cutoff. A tonne diluted from 4 g/t to 3.4 g/t may drop below a 3.5 g/t cutoff and get sent to waste. In the second case, the mine does not recover 85% of the gold; it recovers zero. The relationship between dilution rate and revenue loss in gold is not a smooth curve. It has a cliff at the cutoff grade.

No sensor currently deployed in production can replace the operator's judgment in real time at the ppm grade levels relevant to gold. LIBS detection limits for gold in silicate matrices sit around 5 to 10 ppm (Harmon et al., 2006, Spectrochimica Acta Part B). Below 5 g/t, signal-to-noise is marginal.

The Laubscher cavability chart, originally published in Laubscher (1990, "A geomechanics classification system for the rating of rock mass in mine design," Journal of the South African Institute of Mining and Metallurgy), was developed primarily from massive sulfide and iron ore case histories. Gold deposits with narrow veins and strongly anisotropic jointing often plot in the margins or outside the chart's calibrated domain. Overestimating cavability leads to hang-ups. Hang-up collapse produces air blasts in the drives below.

In tropical high-rainfall climates, saturated fine material above the draw point can liquefy and rush the draw point as a mud inflow. This is a binary hazard: it either happens or it does not, and if it happens there is no time to evacuate. Several gold projects in Indonesia and Papua New Guinea excluded sublevel caving from their evaluations on mud rush risk.

South Africa's Witwatersrand basin is the only gold province where longwall mining operates at scale. The reefs are 0.5 to 1.5 m thick, nearly flat, and 2,000 to 4,000 m deep. There is no other gold deposit on Earth with these characteristics and the technology developed for Witwatersrand is not transferable.

Face advance rate is set by rockburst risk tolerance. At the depths involved, the virgin stress field is close to the rock mass strength. Mining redistributes stress ahead of the face. If the face advances faster than the stress can dissipate through micro-fracturing, energy accumulates and releases as a seismic event. Microseismic monitoring and destress blasting are used to manage this. The safe advance rate for each geotechnical domain was established empirically over decades. McGarr (1999, Pure and Applied Geophysics) and Durrheim (2010, in the proceedings of the 5th International Seminar on Deep and High Stress Mining) discuss the seismological framework.

Room and pillar in gold is a niche application in near-horizontal, moderately thick deposits. The pillar creep issue in altered rock is the main technical concern. Sericite-bearing pillars creep faster than quartzite pillars by roughly an order of magnitude. Delayed failure after a decade or more of loading has caused post-closure subsidence at sites where closure bonds had already been released. The author's familiarity with this method in gold is limited to published case studies and its treatment here reflects that.

Shrinkage stoping is cheap and works in small mines without backfill infrastructure. The sulfide oxidation problem is specific to gold because gold ores almost always carry pyrite or arsenopyrite. Ore stored for months in the stope oxidizes, precipitates iron hydroxides and sulfates, and cements into a solid mass that will not flow. Secondary blasting to free compacted ore in a confined stope is unpleasant work. Acid drainage precursor leaching through the stope floor during storage is a separate liability.

VCR produces high tonnages in thick, steep orebodies. Cumulative blast damage from successive firing rounds weakens the walls of the stope progressively. Overbreak develops in later mining lifts. Magnetic host rocks, which appear in skarn and BIF gold systems, disable magnetic survey tools for hole deviation measurement and require gyroscopic instruments at substantially higher cost. The blast design challenge in VCR on gold is that quartz veins and chlorite-altered wallrock coexist in the same stope and respond differently to explosive energy. Adjusting charge weight hole by hole is routine for experienced blast engineers and difficult for everyone else.

Cut and fill's high powder factor produces finer, more uniform fragmentation than sublevel caving or open stoping. The SAG mill energy draw on fine feed can be 15% to 20% lower than on coarse, heterogeneous feed (e.g., McKee and Jankovic, 2006, in Advances in Comminution, SME). At a 5,000 tpd gold mill where grinding is half the electricity bill, the difference is over a million dollars per year. Nearly every method comparison report treats mill energy as a fixed parameter. It is not. It varies with fragmentation, which varies with the mining method. Mine-to-Mill integration has been standard practice in large copper operations since the early 2000s. Gold has lagged.

Closure costs increasingly enter the method comparison at the front end. Caving methods produce permanent surface subsidence. Backfill methods leave a stable surface. Ontario and Western Australia have raised closure bond requirements. ESG-linked lending covenants are hardening. When the net present value of the long-term environmental liability enters the project economics, the method ranking can shift.

Sensor-based ore sorting underground could change the dilution economics of sublevel caving if the technology matures for gold. XRT and NIR sorting works on minerals that are directly detectable. Gold at ppm levels is not. Sorting has to use arsenopyrite or pyrite as proxy indicators and the strength of the correlation varies by deposit.

Digital twins applied to gold mine planning are better at scenario analysis than at deterministic prediction. A gold orebody with a grade CV over 200% and chaotic jointing will produce different answers from different geological realizations. Running a spread of realizations through each candidate mining method is a way to compare robustness across methods rather than picking a winner on a single model that is probably wrong.

Gold price sensitivity in method selection should be standard practice and mostly is not. Sublevel caving optimized at $1,200 gold looks different at $2,000. At the higher price the dollar value of gold lost to dilution can exceed the cost premium of cut and fill. Open stoping with delayed backfill tends to hold up better across a price range than either extreme because it carries moderate dilution and moderate cost.