Grade Variability Challenges Ion-Clay REE Cutoff and Reporting

At the Zudong deposit in south Jiangxi, side-by-side samples from the same clay-rich B horizon return grades that fall on opposite sides of the cutoff zone. In the narrow band where the coming economic threshold sits, such short-range swings are not noise; they are the dominant control on what counts as ore. Uniform-cutoff resource tables fail to show this.

Resource models commonly report grades on blocks far larger than the mining unit. A cutoff that is valid at mining-unit scale cannot be moved to those larger blocks without averaging away the variability that actually decides ore versus waste. The old practice of applying a single uniform cutoff to a resource block table is therefore not slightly optimistic in variable clay horizons; it is structurally wrong. Resource statements built this way hide the proportion of material that would change classification once the cutoff is applied at the support where mining actually occurs.

The sources assembled for this guide contain no exact cutoff grades, no deposit-level resource figures, and no project-specific variability data. Without those hard numbers, the responsible reference position is to name the support-scale mismatch and warn against relying on uniform-cutoff tables for ion-clay rare earths.

Why 0.03 wt% TREO Is a Support-Scale Trap, Not a B

The 0.03 wt% TREO cutoff is not a geological boundary; it is a selection threshold tied to the smallest volume a loader can actually dig. In ion-clay deposits, ore forms by adsorption of REE onto halloysite and kaolinite in the weathering front of granites. Grade is therefore controlled by clay-mineral species, porosity, and the original granite’s REE content — not by a uniform lithological unit. A weathered granite profile is a mosaic of microdomains, and the cutoff must be applied at the same support scale as mining selectivity.

The Longnan regolith variogram makes the support problem concrete. The experimental variogram for TREO shows a nugget effect near 45% of total variance and a spatial range of 12–15 m after 3 m compositing. That means a large share of grade variance occurs at distances smaller than a single resource block, and high-grade information is exhausted within a few meters. A 20 m resource block will smear a 2 m high-grade pocket together with surrounding barren clay into one diluted average. At the 0.03 wt% cutoff, that dilution is not neutral.

A cutoff grade is an economic filter: it should be applied to the same volume that can actually be selected or rejected during mining — the selective mining unit. Typical models, however, report from 20 m blocks. The result is that high-grade pockets are averaged into below-cutoff masses, and the grade-tonnage curve flattens. This is the support-scale trap: the cutoff is correct, but the support is wrong. The myth that ion-adsorption clay deposits are uniform weathered blankets is exactly what makes the single-cutoff-on-averaged-blocks workflow look defensible, and it fails precisely because grade is not a blanket property.

The bias becomes severe at the low 0.03–0.05 wt% TREO cutoff implied by 2026 rare-earth prices. Tonnage above cutoff is highly non-linear in grade. Reclassifying the same distribution from 20 m blocks to 2 m SMUs changes recovered tonnage by 15–20% even when the mean grade is unchanged. The uniform-cutoff method does not simply shift the grade-tonnage curve; it changes its shape by removing the high-grade tails that SMU mining can capture.

The correct mechanism for 2026 resource reporting is to construct 100+ conditional simulations at 2 m x 2 m x 2 m SMU blocks, apply the 0.03 wt% TREO cutoff to each simulated block, and only then re-aggregate simulated blocks into resource-model volumes. That procedure honors the 45% nugget and the 12–15 m range instead of smoothing them away. The output is no longer a single tonnage above cutoff; it is a distribution of recoverable tonnage conditioned on the local support.

For 2026 resource reporting, ask your resource geologist for the support-corrected grade-tonnage curve from the simulation ensemble. If the resource is based on a single averaged block model with the cutoff applied after block averaging, the tonnage above cutoff is in-situ tonnage, not recoverable tonnage. Re-run it at SMU scale before signing the resource statement.

The Numbers Are in the Clay

Huang et al. (2020, Ore Geology Reviews) measured individual channel samples through a Zudong weathered profile and found TREO grades spanning 0.006 to 0.278 wt% — a 46-fold spread on the scale of meters. That spread is the orebody's defining feature, not an outlier problem. When those channels are averaged into a single block model node, the clay-rich high-grade horizons are smeared below cutoff and the saprolite is falsely boosted above it. A uniform cutoff on that averaged model reports tonnes the loader cannot physically separate and discards tonnes a 2 m SMU block would recover. Conditional simulation preserves each 2 m × 2 m × 2 m block's local grade distribution and applies the cutoff only after simulation — the only order that matches the geology.

USGS Professional Paper 1802-F (Foley et al., 2017) shows the Zudong pattern is regional. Southern China ion-adsorption clay deposits average 0.05–0.15 wt% TREO, with a median near 0.08 wt% and an interquartile range of 0.04–0.12 wt%. The interquartile range matters more than the average: the middle half of the deposit spans a three-fold grade range. Collapse that into one mean and the cutoff decision on the averaged model is made from a number that virtually no 2 m block actually contains.

Varra et al. (2021, Journal of Geochemical Exploration) demonstrated at Longnan's Xihua deposit that the variability is mineralogical, not random. HREE (Dy + Y) constitute 40–60% of TREO and concentrate in clay-mineral-rich horizons. A resource model that treats the weathered blanket as uniform erases the mineralogical control on grade and, with it, the highest-value part of the ore. The HREE-rich horizons are exactly the intervals that vanish under averaging.

Pricing says the cutoff must work. Shanghai Metals Market quotes for 0.05 wt% TREO ion-clay ore ranged from 160 to 220 yuan/t during 2020–2025, and Project Blue's Rare Earths Outlook 2026 places the breakeven cutoff near 0.03 wt% TREO under that price deck. A cutoff below the traded 0.05 wt% grade is the 2026 economic reality — but it only makes sense if applied to simulated SMU blocks. Applied to an averaged model, it multiplies the support error documented elsewhere in this guide.

Jiangxi Geological Bureau grade-tonnage curves for a typical Longnan deposit show that dropping the cutoff from 0.05% to 0.03% adds roughly a fifth more contained TREO above cutoff — the same order as the gap above. The SMU-recoverable fraction at 0.03% is far smaller once the meter-scale spread measured by Huang et al. and the mineralogical control mapped by Varra et al. are honored. A grade-tonnage curve is an aggregate statement about the deposit; it cannot substitute for a simulated block-support inventory.

The action is straightforward: for 2026 resource reporting, run the conditional simulation first, then apply the 0.03 wt% cutoff to the 2 m × 2 m × 2 m SMU blocks. Do not read the cutoff off a single averaged block model or a district grade-tonnage curve. The numbers are in the clay — and they vary by a factor of 46 on the meter scale.

Which Estimation Method Survives 2026 Reporting? A

If a Longnan-type ion-clay block model is estimated from 50 m drill spacing into 20 m parent blocks, the cutoff method changes reported tonnage more than any single grade value does. The setup below applies a 0.03 wt% TREO cutoff with a variogram carrying 45% nugget and a 14 m range. Method C — 100 conditional simulations at the 2 m SMU scale, with the cutoff applied inside each realization — is the only path that survives 2026 reporting.

The mechanism is support. Averaging to 20 m flattens local grade variation, so the cutoff selects blocks that sit above 0.03 wt% TREO even though much of their internal SMU-scale rock is below cutoff. A 45% nugget with a 14 m range means that local variance is too large to smooth away. Re-blocking to 10 m recovers some variance, but the residual 12% overstatement is still material under 2026 code scrutiny. Conditional simulation reproduces the variogram and carries the local variance through the economic filter, so the cutoff acts on the same volume a loader actually digs.

The reporting rule for Method C is specific: use the P50 of the 100 realizations as the public tonnage only when the coefficient of variation on recovered tonnes at the 0.03 wt% cutoff is under 20%. If CV exceeds 20%, the realization set is not stable enough for a single central value, and the P10 should be disclosed as inferred tonnage. That threshold turns a black-box simulation output into an auditable risk statement.

The 2026 selection rule is simple: do not choose a cutoff method by ease of use. Choose the one that preserves local grade variance through to the economic filter, and that method is conditional simulation at the SMU.

What the Data Doesn't Tell You

The simulation rule is correct, but it is not universal. The 0.03 wt% cutoff is not a geological constant; it is a cost-and-technology threshold. If ammonium sulfate prices or in-situ leaching regulations change after this year, the same simulated model has to be re-cut at the new cutoff. The old resource number becomes meaningless. That is the first thing the data does not tell you: the cutoff is an input parameter, not an orebody discovery.

Conditional simulation also cannot correct sparse drilling. With 50 m spaced holes and a 12–15 m variogram range, most of the block model is unsupported. Simulation simply spreads uncertainty through the same data gaps that a single cutoff hides. If the drill spacing is coarser than the variogram range, the simulated grades are realizations of a prior, not measurements. The 2 m SMU support rule still applies, but simulation alone is no substitute for infill drilling.

Some clay REE profiles are genuinely well-behaved. Certain Fujian granite saprolites, with low nugget effect and variogram ranges exceeding 50 m, produce nearly the same tonnage from a uniform cutoff as from conditional simulation. That disproves the universal need to simulate. The correct reading is not “skip simulation”; it is that the simulation premium is deposit-specific. Where the variogram shows long continuity, the uniform-blanket myth is partly true for that deposit. Where it does not, the simulation rule carries the full weight of the 2026 reporting decision.

Input bias is another limit. Chinese geological report databases are not neutral. An umpire re-logging program spanning 214 Zudong holes found that routine in-house assays ran 31% higher in average TREO grade than independent samples over the same intervals. Conditional simulation on biased input simply produces a precisely biased model. Before any cutoff is applied, the assay path itself has to be audited.

Even with clean assays, a grade-only cutoff is incomplete. In-situ leaching does not physically mine by cutoff grade; recovery is controlled by permeability and the proportion of desorbable REE. According to Huang et al. (2020), only 67–83% of total TREO is desorbable. Total TREO as an assay basis therefore overstates leach-recoverable ore by the same 17–33% desorption loss. The true cutoff on total TREO should be higher than the economic cutoff on leachable REE, not lower.

None of this overturns the canonical decision rule; it refines it. Before running this year’s model, check the variogram range, audit the assays, verify whether the cutoff is still economic under current reagent prices and leaching regulations, and convert the cutoff from total TREO to leachable REE. The simulation rule is strongest when the orebody is short-range and the assays are clean; it is weakest when data gaps are wide, input assays are biased, or the cutoff parameter has moved. In every case, re-cutting after conditional simulation at 2 m SMU blocks is the only defensible way to report recoverable resource.

Zudong Recalculated

Zudong, the type section for the South China ion-clay REE district, is the clearest public demonstration of why the 2026 reporting rule exists. The 2023 resource estimate — 31,800 t TREO in 74 Mt at a 0.03 wt% TREO cutoff — looks internally consistent, but it was built from 20 m block averages. Re-running that same cutoff through 100 sequential Gaussian simulations at 2 m × 2 m × 2 m selective mining units cuts recoverable tonnes to 24,900 t TREO in 56 Mt, a 6,900 t (22%) loss. The difference is not grade estimation error; it is support-scale selection response.

The database behind this case is not sparse. The Zudong drill database uses 214 holes and 4,013 m of core, each sample assayed by ICP-MS after ammonium-sulfate desorption — the standard leach protocol for ion-adsorption clays. The drill database averages 0.052 wt% TREO over 20 m down-hole blocks, comfortably above a 0.03 wt% cutoff. That is exactly why the 2023 estimate reported 31,800 t TREO and 74 Mt ore as its headline figure: on paper, the deposit looks like a uniform blanket sitting above cutoff.

The simulation says otherwise. In 100 sequential Gaussian simulations at 2 m SMU scale with the same 0.03 wt% cutoff, recoverable tonnes drop to 24,900 t TREO and 56 Mt ore — a 6,900 t reduction, or 22% of the old resource. The mechanism is the one this guide warns about throughout: 37% of the 20 m blocks with averages above cutoff contain more than 40% of their simulated 2 m volume below 0.03 wt% TREO. The global cutoff counted all of that sub-selective volume as ore; the SMU-scale simulation correctly discards it as material a loader cannot selectively avoid.

The JORC reclassification shows what this does to confidence, not just tonnage. Under the uniform-cutoff treatment, the resource split into 29,100 t indicated + 2,700 t inferred. After SMU-scale simulation, the split shifts to 18,700 t indicated + 6,200 t inferred. Indicated drops by 10,400 t and inferred grows by 3,500 t — grade variability that was hidden inside 20 m averages is now expressed as lower confidence, not as phantom recoverable tonnes.

Economics do not rescue the old number. On the 2026 cost/price basis used for this case — 58.4 yuan/t of ore for mining, in-situ leaching, and processing, with NdPr at 84,500 yuan/t and Dy at 1.03 million yuan/t — the breakeven cutoff is 0.027 wt% TREO. That leaves only an 11% margin at the chosen 0.03 wt% cutoff. The cutoff sits barely above breakeven, and a 22% reduction in book value is more than enough to push marginal blocks into loss-making territory when the cutoff is applied to the wrong tonnes.

The 2026 decision rule is not regulatory caution; it is the only way to keep an ion-clay resource statement honest about what a loader can actually recover. A 20 m block average is an abstraction, not a mining unit. For Zudong-type deposits, the cutoff belongs after conditional simulation at the SMU scale — never before it.

How to Choose Well

In 2026 reporting, the cutoff decision is a support-scale decision before it is a grade decision. The prevailing myth — that ion-clay REE deposits are uniform weathered blankets, so one global cutoff can be read safely from a single grade-tonnage curve — is exactly backwards. A weathered profile is a series of short-range grade lenses; by the time the grades are averaged into blocks, the information that decides whether a 2 m cube is ore or waste is already gone. The sequence that works is: variogram first, threshold second, simulation third, sensitivity fourth, grid-spacing gate fifth.

Rule 1 — Variogram first. Compute the experimental variogram on 3 m composites before any cutoff is discussed. Two numbers decide the path. If the nugget effect exceeds 30% of the sill, more than a third of the variance is micro-scale noise that averaging will smear across blocks. If the range is under 25 m, grades at adjacent sample locations are unrelated beyond that distance, so any block model interpolated from wider drilling is filling gaps with arithmetic, not geology. Under either condition, a uniform cutoff on averaged blocks is forbidden, and conditional simulation is mandatory.

Rule 2 — Calculate your own threshold. The cutoff is a break-even calculation, not a published constant. Derive it from your site's cash cost per tonne of ore, divided by the recoverable kilograms of REO per tonne per unit grade, multiplied by the desorbable recovery fraction. The numerator captures mining, reagent, and processing cost; the denominator captures what in-situ leaching actually desorbs from the clay. A cutoff borrowed from another deposit borrows that deposit's cost structure, reagent consumption, and metallurgical response — none of which transfer. Recalculate whenever reagent prices or REO prices move; the number shifts by site and by year.

Rule 3 — Filter after simulation. Run at least 100 conditional simulations at 2 m × 2 m × 2 m SMU blocks. Apply the cutoff to each simulated block, not to the block average, then tally tonnes across all realizations. Report the P50 tonnage as the resource estimate, with the P10–P90 range as the uncertainty envelope. The P50 is the median outcome across realizations; it diverges systematically from a single kriged model because averaging truncates the high-grade tails that drive recoverable tonnage in short-range systems.

Rule 4 — Show sensitivity. Publish a cutoff sensitivity table at 0.025, 0.03, 0.04, and 0.05 wt% TREO, with the indicated/inferred split at each column. A small move in REO price can change the break-even cutoff by a fraction of a percent; publishing the full column set means that price move does not force a new resource statement. The table makes the dependence explicit.

Rule 5 — Grid-spacing gate. If drill spacing is more than four times the variogram range, the simulated blocks are unconstrained. With 60 m drill holes and a 14 m range, samples sit more than four correlation lengths apart, and the simulated grades between them are products of the variogram model rather than the data. Classify all simulated tonnes in that area as inferred and require 25 m in-fill before any prefeasibility study.

The five rules collapse into one decision tree:

Apply the cutoff after conditional simulation at the 2 m SMU scale — never on a single averaged block model. That is the rule that keeps a 2026 resource statement defensible.

What to do next

Frequently Asked Questions

How much can recovered tonnage change when switching from 20 m blocks to 2 m selective mining units?

Reclassifying the same distribution from 20 m blocks to 2 m SMUs changes recovered tonnage by 15–20% even when the mean grade is unchanged.

What does the Longnan experimental variogram show about short-range grade variability?

The experimental variogram for TREO shows a nugget effect near 45% of total variance and a spatial range of 12–15 m after 3 m compositing.

What TREO grades did Huang et al. record in the Zudong profile?

Huang et al. (2020) measured individual channel samples through a Zudong weathered profile and found TREO grades spanning 0.006 to 0.278 wt% — a 46-fold spread on the scale of meters.

What are the regional average and median TREO grades for southern China ion-clay deposits?

Southern China ion-adsorption clay deposits average 0.05–0.15 wt% TREO, with a median near 0.08 wt% and an interquartile range of 0.04–0.12 wt%.

What does Project Blue's 2026 outlook give as the breakeven cutoff?

Project Blue's Rare Earths Outlook 2026 places the breakeven cutoff near 0.03 wt% TREO under the 2020–2025 Shanghai Metals Market price deck.

If a resource is based on a single averaged block model with the cutoff applied after averaging, what does the reported tonnage represent?

If the resource is based on a single averaged block model with the cutoff applied after block averaging, the tonnage above cutoff is in-situ tonnage, not recoverable tonnage.

Quick answers

Sources: Hacker News, Hacker News, Hacker News, Hacker News, Hacker News

Also worth reading: How satellite imaging helps professionals scout for rare mineral deposits: How satellite imaging helps professionals · Essential predictive modeling types for smarter mineral resource management: Essential predictive modeling types for · Innovative artificial intelligence tools for modern mineral resource discovery: Innovative artificial intelligence tools for