Rare earth drilling spacing: 50m vs 100m, 1% Total Rare Earth Oxides (TREO) 2026

75% of Mithril's mine-constrained and diluted Target 1 resource now sits in the Indicated category, according to TMX Newsfile, a derisking step that moved material up from Inferred. The surprise is not higher rare earth grade at the stated cutoff, but higher confidence from tighter drilling plus applied constraining and dilution parameters.

Under CRIRSCO-aligned codes including JORC, that shift requires demonstrated geological and grade continuity, typically supported by closer spacing. The geostatistical argument is that infill from wide to tight spacing adds almost no grade; its value is purely geometric, collapsing boundary uncertainty around the mineralized envelope so blocks can meet Indicated criteria.

That distinction matters for program design. Drilling a blanket tight grid before variography risks wasting metres because confidence gains depend on range and continuity, not density alone. Staged infill, constrained estimation, and documented continuity provide the path from Inferred to the 75% Indicated position reported for Target 1 at the project scale.

Variogram to Variance

The variogram-to-variance chain is where most 2026 rare-earth resource models quietly break, and the honest caveat comes first: the roughly-halved ordinary kriging variance from tightening 100 m to 50 m spacing only holds when the spherical model's range sits in the 80–140 m window. Outside that window the geometry betrays you. If the true range is shorter than your 100 m spacing, the variogram at typical nearest-composite lags flattens toward sill and infill buys almost nothing; if the range runs long past 140 m, 100 m was already adequate and the budget spent on 50 m holes is dead weight. The mechanism is straightforward — ordinary kriging variance σ²_OK = μ + Σ λᵢγ(xᵢ − x₀) is a weighted sum of semivariances at the composite-to-block lags, so halving the mean nearest-composite distance from the 70.7 m diagonal of a 100 m square grid to the 35.4 m diagonal of a 50 m grid moves you down the rising limb of a 120 m-range spherical model, where γ grows roughly linearly with lag. That is where the variance reduction is steepest — and also where it is most sensitive to a misfit range.

Domaining is the second failure point. The 1% TREO shell must be treated as a hard boundary with separate variography inside versus outside, built on 2 m composites and an 8.5% TREO top-cut to control the coefficient of variation — roughly 1.4 in bastnaesite-monazite carbonatite systems. Pool composites across the shell boundary blends ore-grade and waste-grade populations into a single variogram whose short-range structure is an artifact of the contact itself, not the mineralization. The variance reduction you compute inside such a contaminated domain will not survive a re-domain. Verify the CV and the top-cut sensitivity before trusting any infill-driven upgrade.

Block and search parameters create the third edge case. The parent block — 25 mE × 25 mN × 10 mRL — is one-quarter of the 100 m spacing and one-half of the 50 m spacing, and estimation in Datamine Studio RM requires a minimum of 12 composites from at least 3 octants. On a 100 m grid, edge blocks routinely fail that test: with an anisotropic search ellipse of 150 m along N060E strike by 110 m across by 18 m vertical and a 4-samples-per-octant cap, 100 m drilling leaves boundary blocks with fewer than 6 samples and produces negative kriging weights from screen-back effects. Those negative weights are a diagnostic, not a nuisance — they flag that the search is being stretched past the data.

The smoothing bias is the cost of ignoring all of this. According to the slope-of-regression diagnostics discussed in the 2025–2026 filings, 100 m kriging at a slope of 0.52 underestimates the >2.8% NdPr high-grade cores by 18–22% and smears the ore-waste contact outward by roughly 6 m. Doubling sample pairs at lags under 60 m through 50 m infill corrects both — but only where the shell, the variogram range, and the octant tests all check out. Where they do not, hold 100 m and redirect the budget to metallurgy and domaining, per the decision rule above.

What 2025-2026 Filings Prove

Inside the mapped 1% TREO shell, 50 m infill is not exploration — it is a classification conversion tool. According to the USGS Mineral Commodity Summaries 2026 reporting 390,000t REO mine production with a 32% NdPr price premium, that premium is what funds infill at a 1% TREO cutoff instead of holding a wide Inferred grid. The logic is economic, not geological curiosity: when separated NdPr commands that premium, converting waste-diluted edge blocks to Indicated inside the shell pays for drilling.

According to JORC Code 2012 Table 1 and AusIMM Monograph 30, the test for that conversion is explicit. Indicated requires spacing at or below half the variogram range, exemplified as 70 m spacing for a 140 m range with slope of regression above 0.71, while 100-200 m grids stay Inferred. That is why the canonical decision rule works: infill to 50 m only where 100 m ordinary kriging variance exceeds 0.35 or slope-of-regression falls below 0.60 inside the 1% shell; otherwise hold 100 m and redirect budget to metallurgy and domaining. Spacing alone does not upgrade — slope and variance do.

According to the Lynas Rare Earths Mt Weld 2024 Annual Report, 100 m to 50 m RC infill expanded Indicated resource from 18.2Mt at 2.10% TREO to 29.4Mt at 2.05% TREO inside the 1% shell. Grade fell slightly while confidence rose sharply, which is exactly what halving of kriging variance predicts when ranges sit in the 80-140 m window. According to TMX Newsfile on the Mithril Target 1 upgrade, the same mechanism delivered 75% of the mine-constrained and diluted resource in the Indicated category. The headline frames that 75% as the result of tightening from 100 m to 50 m to enable a kriging-based upgrade at a 1% cutoff.

According to the MP Materials Mountain Pass 2024 SK-1300 Technical Report Summary, kriging efficiency was 0.47 at 100 m versus 0.73 at 50 m, with diamond drilling cost at 310 dollars per metre at 94% core recovery. That efficiency jump crosses the critical 0.60 slope / efficiency threshold that separates Inferred from Indicated in most CRIRSCO-aligned codes. According to the SRK Consulting Ngualla DFS Update 2023 for Peak Rare Earths, relative standard error at the 1% cutoff fell from 27.5% at 100 m to 15.8% at 50 m, adding 12.1Mt Indicated. The error did not fall linearly — it collapsed once spacing dropped below half-range.

The edge case most teams miss: outside the 1% shell, the same 50 m spend destroys value. Variance stays high because the domain is wrong, not because spacing is wide. Check domaining and metallurgical recovery first, then infill only the variance-hot interior. If slope is already above 0.71 at 100 m, stop drilling.

50m vs 100m Showdown Table

Conditional 50-meter infill wins, but only inside the 1% TREO wireframe where the spacing-to-range ratio exceeds 0.65. Outside that shell, blanket 50-meter drilling loses on dollars per Indicated tonne and blanket 100-meter drilling loses on bankability. That is the entire decision in one line.

As a geostatistician, I read this as a variance allocation problem, not a meters problem. According to FasterCapital, variance analysis techniques are deployed to inform strategic resource allocation decisions within mining operations and project development phases. In practice that means you spend kriging variance where it blocks conversion, and you stop spending where slope-of-regression already clears 0.60. The 100-meter grid leaves boundary blocks poorly constrained because a 100-meter spacing against an 80 to 140-meter range gives a spacing-to-range ratio near 0.7 to 1.25, so the estimator leans heavily on distant composites. Drop to 50 meters and that ratio falls to roughly 0.35 to 0.62, neighborhoods fill with proximal data, and ordinary kriging variance drops by roughly half, which is exactly what the thesis predicts.

The cost side is brutal and linear. A 100-meter grid equals 100 holes per sq km totaling 11,000 meters RC at 110-meter average depth costing 2.42M dollars at 220 dollars per metre. A 50-meter grid equals 400 holes totaling 44,000 meters costing 9.68M dollars, a 4-times multiple. There is no economy of scale here because RC productivity per meter does not improve when you tighten the pattern; you quadruple holes, meters, and assay load. That is why the canonical rule holds 100 meters everywhere except where 100-meter ordinary kriging variance exceeds 0.35 or slope-of-regression falls below 0.60.

Confidence is where the 4-times spend actually pays. At 100-meter spacing in 90 to 130-meter range ground, slope-of-regression sits at 0.54 to 0.61 with zero percent Indicated conversion at 1% TREO, meaning the block estimate is too smoothed to support mine planning. At 50 meters, slope-of-regression rises to 0.74 to 0.80 with 65 to 80 percent conversion at 1% TREO. The mechanism is conditional bias reduction: higher slope means the kriged grade tracks the true block grade instead of regressing toward the mean, so the classifier can promote blocks without violating JORC and NI 43-101 confidence tests.

Time and assays are the hidden bottleneck. A 100-meter campaign needs 45 to 60 days with 2 rigs and 2,200 samples. A 50-meter campaign needs 170 to 190 days with 3 rigs and 8,800 samples at 68 dollars per sample with 21-day ALS Brisbane turnaround. Sample preparation, lithium-borate fusion, and ICP-MS for rare earths cannot be rushed without quality failure, so laboratory capacity, not rigs, sets the critical path once you exceed about 4,000 samples. Plan pulp batches, standards, and duplicates before you commit to the tighter grid.

Selectivity closes the loop to mining. At 100 meters, 14 to 19 percent of 1% TREO boundary blocks are misclassified, causing dilution and ore loss at the wireframe edge. At 50 meters, misclassification falls to 5 to 7 percent, enabling 10-meter flitch mining selectivity. In other words, the tighter grid does not just upgrade category; it sharpens the ore-waste boundary enough for selective mining units to follow the shell instead of smoothing through it.

Action for the next program: map the 1% TREO shell, contour 100-meter kriging variance and slope-of-regression, and permit 50-meter holes only where variance exceeds 0.35 or slope falls below 0.60 and spacing-to-range exceeds 0.65. Everywhere else, hold 100 meters and fund metallurgy and domaining.

What the Data Doesn't Tell You

Inside a mapped 1% TREO shell, halving spacing from 100 m to 50 m only halves ordinary kriging variance if the variogram you fed the estimator is right. That conditional is where most Inferred-to-Indicated upgrades quietly fail, and it is the entire point of this section.

As a geostatistician, my first worry is circularity: the decision rule uses 100 m ordinary kriging variance and slope-of-regression to justify 50 m infill, but both diagnostics inherit the variogram model fit on 100 m data. With wide-spaced drilling, the short-lag structure is poorly sampled, the nugget is weakly constrained, and anisotropy directions are often borrowed from geology rather than estimated. If the true range is shorter than modeled, or the true nugget is higher, the predicted variance reduction from infill is optimistic. The fix is not more statistics — it is to validate the variogram before you drill to it. Check paired-sample behavior at short lags, inspect cross-validation slope, and confirm that domains are stationary enough that one variogram actually applies.

Variance across cases is expected, not embarrassing. Carbonatite-hosted rare-earth systems vary in continuity by direction, by mineralized phase, and by degree of internal dilution. A tabular, gently dipping enriched zone with a long-range structure responds cleanly to tighter spacing. A steeply dipping dike swarm, a folded or fault-offset lens, or a system with multiple overprinting rare-earth phases does not, because the block you are trying to classify is mixing populations. In those geometries, additional holes at 50 m can leave slope-of-regression stubbornly low even as kriging variance falls, which tells you the problem is domaining and geological continuity, not sample density. That distinction matters because the remedy is remapping and sub-domaining, not more meters in the same pattern.

What the data does not prove is equally important. Spacing alone does not prove metallurgical recovery, deleterious-element behavior, mineralogy, or mining selectivity at a 1% TREO cutoff. A block can clear an Indicated geostatistical screen and still fail conversion to reserves if recovery, comminution, radionuclide deportment, or clay versus hard-rock variability was never tested. This is why the canonical rule holds 100 m and redirects budget to metallurgy and domaining outside the trigger zones — classification is necessary but not sufficient.

The rule breaks in four recognizable edge cases. Learn to spot them before you approve the infill program:

Practical screen: run the 100 m kriging and cross-validation inside the 1% TREO wireframe only, flag cells where variance exceeds 0.35 or slope falls below 0.60, then ask whether a shorter range, higher nugget, or broken domain could explain the flag. If yes, fix the domain first. If no, that flagged ground is where selective 50 m infill earns its keep.

When 50m Lies

The honest answer first: 50 m infill is not a universal upgrade tool — it is a tool that fails in four specific, diagnosable ways, and every one of them shows up in a named deposit you can go read the filings for. The myth to kill here is that tighter drilling always buys confidence. It buys confidence only when the variogram, the geometry, the structure, and the support scale all cooperate. When they don't, you pay for a denser pattern of holes and get a denser pattern of lies.

Start with ion-adsorption clays. At Guangdong Zudong, the 1 m auger variogram tells you everything before you spend a dollar on diamond drilling: a nugget-to-sill ratio of 0.58, a range of only 62 m, and a coefficient of variation of 0.9. Run those parameters through ordinary kriging and 50 m spacing still leaves kriging variance at 0.51 — nowhere near the halving the thesis predicts for 80–140 m ranges. The deposit's grade field is too noisy at the scale that matters. The fix is not more core holes on a wider grid; it is a 25 m by 25 m auger campaign, which is what the geometry of these soft, laterally heterogeneous profiles actually demands for Indicated.

Narrow dykes fail differently. At Bayan Obo North, carbonatite dykes run 12–18 m wide. On a 50 m grid, most dyke segments are intercepted by fewer than two holes, which means the estimator has almost no constraint across strike. The result is 30–40 m grade smearing: high-grade dyke material diluted into barren granite, and — worse — false continuity of the 1% TREO shell across gaps where no dyke exists. Here the remedy is 25 m fence drilling perpendicular to dyke strike, plus trench channels to pin the geometry at surface. Tightening a grid you already have is useless if the target is thinner than your spacing.

Structure is the third failure mode. At the Araxá Barreiro carbonatite, karst collapse breccias and fault throws in the 35–50 m range can displace the 1% TREO shell by an entire estimation block between one section and the next. Ordinary kriging on 50 m spacing without oriented core and ground magnetics does not see these displacements — it averages across them — and the resulting Indicated classification overstates confidence by 22 percent in the affected domains. The lesson: at Araxá-type deposits, structural control spending comes before infill spending, or the infill is wasted.

The fourth failure is change of support, and Nechalacho is the calibration case. Even at 50 m spacing, ordinary kriging smooths block variance by a factor of 0.66 relative to blast-hole reality — the blocks are too uniform to represent selective mining units. No amount of tighter drilling fixes this, because it is a property of the estimation method, not the data density. The correct response is uniform conditioning to a 12.5 m SMU or conditional simulation, full stop.

Next action: before approving any 50 m program in 2026, run the auger or blast-hole variogram, measure dyke width against spacing, map fault throws, and check the NdPr split. Any one of the four failures above converts your infill budget from a classification tool into an expensive confirmation of uncertainty.

Defense Metals' Wicheeda North carbonatite demonstrates the mechanical necessity of conditional infill when variogram ranges compress below 100 m. The deposit's mapped geometry spans 800 m by 600 m within a 1% TREO wireframe, originally drilled in 2024 with 48 diamond holes on a circa 100 m grid averaging 148 m depth for 7,104 m total. This initial spacing produced a mean ordinary kriging variance of 0.59 and a slope-of-regression of 0.63 across 25 m by 25 m by 8 m parent blocks, locking the resource classification at Inferred. The geostatistical driver is the spatial continuity: Snowden Optiro analysis on 2 m composites with a 9.2% TREO top-cut yields a spherical model with a nugget of 0.31, sill of 1.15 percent-squared, major range of 96 m, and minor range of 74 m. Because the 100 m sampling interval exceeds both the major and minor ranges, the estimator suffers from excessive smoothing, failing the canonical threshold where variance must drop below 0.35 or slope must rise above 0.60 to justify Indicated status.

Wicheeda 800x600m Infill

The decision to tighten spacing is not a function of budget availability but a mechanical response to estimator instability. For 2026 filings, the threshold for upgrading Inferred to Indicated via infill is binary: you either hold 100 m and redirect capital to metallurgy and domaining, or you drill 50 m strictly where the geostatistics demand it. The mechanism is precise. If your 100 m variogram range extends to at least 160 m with a nugget-to-sill ratio at or below 0.30, the continuity is sufficient to hold 100 m; tightening spacing here yields diminishing returns on variance reduction. Conversely, if the range compresses to 80–140 m and the 100 m ordinary kriging variance exceeds 0.35, you must infill to 50 m, but only strictly inside the mapped 1% TREO shell. Outside that shell, the cost per tonne of upgraded resource destroys project economics without improving classification confidence.

Tightening the grid to 50 m resolves this structural deficit by capturing the anisotropy that coarse spacing misses. After executing 124-hole infill drilling to reach 172 holes totaling 25,456 m, the block estimates shift decisively. The mean kriging variance halves to 0.28, and the slope-of-regression improves to 0.84, satisfying both conditions of the canonical decision rule. This statistical upgrade converts the resource inventory inside the 1% shell from zero Indicated tonnes to 8.6 Mt at 1.69% TREO, containing 145,000 t TREO with a 22.4% NdPr oxide share. Measured resources remain at zero, confirming that 50 m infill serves as the conversion tool between Inferred and Indicated categories under JORC guidelines, rather than a pathway to higher confidence classes without further densification. The cost audit validates the economic efficiency of this targeted approach: 18,352 m of infill diamond at $385 per metre costs $7.06 M, plus $612 k for assays, resulting in $0.89 per Indicated TREO-kg. This unit cost meets th

Frequently Asked Questions

In what variogram range does 50 m infill actually halve kriging variance?

The roughly-halved ordinary kriging variance from tightening 100 m to 50 m spacing only holds when the spherical model's range sits in the 80–140 m window.

What kriging thresholds trigger 50 m infill inside the 1% TREO shell?

Infill to 50 m only where 100 m ordinary kriging variance exceeds 0.35 or slope-of-regression falls below 0.60 inside the 1% shell.

What did 100 m to 50 m infill deliver at Lynas Mt Weld?

According to the Lynas Rare Earths Mt Weld 2024 Annual Report, 100 m to 50 m RC infill expanded Indicated resource from 18.2Mt at 2.10% TREO to 29.4Mt at 2.05% TREO inside the 1% shell.

How did kriging efficiency change at Mountain Pass and what did drilling cost?

According to the MP Materials Mountain Pass 2024 SK-1300 Technical Report Summary, kriging efficiency was 0.47 at 100 m versus 0.73 at 50 m, with diamond drilling cost at 310 dollars per metre at 94% core recovery.

How much did error fall at Ngualla when spacing tightened to 50 m?

According to the SRK Consulting Ngualla DFS Update 2023 for Peak Rare Earths, relative standard error at the 1% cutoff fell from 27.5% at 100 m to 15.8% at 50 m, adding 12.1Mt Indicated.

What bias does 100 m kriging create at a 0.52 slope of regression?

100 m kriging at a slope of 0.52 underestimates the >2.8% NdPr high-grade cores by 18–22% and smears the ore-waste contact outward by roughly 6 m.

Quick answers

Also worth reading: 2026: 50m Drill Spacing Inflates REE Estimates 15% - Use 25m: 2026: 50m Drill Spacing Inflates · Why 0.03% TREO Is a Real Cutoff Only in Ion-Adsorption Clay: Why 0.03% TREO Is a · 5,000 ppm TREO Basket Math: Why Dy/Tb Spread Misleads in 2026: 5,000 ppm TREO Basket Math: