Critical Materials Atlas
Method · cross-source comparison

Two counts of the same world

Every production figure the atlas publishes comes from a compilation of what countries report. So we ran the obvious check: does a second public compilation say the same thing? For 35 of 53 materials the two track within 10%. The useful part is the ones that don’t — because each disagreement turns out to name something specific about how the world’s mineral statistics are actually made.

We compare the sum of BGS World Mineral Statistics national returns against the USGS world-production estimate for the same material, over the years both cover. These are independent compilations, but not independent measurements — both rest largely on the same national statistical returns, so they can agree and still both be wrong. Agreement therefore tells you we have paired the right forms and read the units correctly. It is not evidence that a number is true, and a disagreement does not make either body wrong. Most disagreement here is definitional.

The disagreements are the content

One was unexplained until recently, and it turned out not to be an error in anyone’s data. A second, cobalt, was “explained” here and has been withdrawn: the gap was our own double count, not the data (see below).

Cementratio 0.07China and India are absent

BGS cement has ~32 reporting countries, led by Turkey, Germany, Italy and Spain. China alone produces roughly 2.4 of the world’s ~4.2 billion tonnes, and it does not appear at all. The panel is a Europe-weighted subset, not a world census — so its sum must never be used as a world denominator. Nothing here is a data error; the series simply does not mean what a reader would assume it means.

The limit that outranks the ratio. A sum over a handful of reporting countries cannot be a world census, however neatly the ratio lands. 13 of 53 materials here rest on fewer than eight reporting countries — including lithium and rare earths. Germanium is the clearest case: its two figures happen to sit close together, but BGS carries only three reporting countries for it, so that agreement is coincidence rather than confirmation. Those rows are marked † below, and the mark should be read before the number.

All 53 materials

Ordered by how far apart the two counts are — least comparable first, because those rows carry the information. Ratio = median (BGS national sum ÷ USGS world estimate) over the overlapping years.

MaterialBGS form pairedRatioReportersStatusWhat the difference is
tantalum †tantalum and niobium minerals122.806not comparableBGS publishes tantalum AND niobium minerals as one gross-ore series; USGS publishes contained tantalum alone. Different numerator and different basis.
beryllium †beryl24.986not comparableBGS counts BERYL (the ore mineral); USGS counts contained beryllium. Beryl is roughly 4% Be.
chromiumchromium ores and concentrates2.8219not comparableBGS counts chromite ORE at gross weight; USGS counts contained chromium.
manganesemanganese ore2.7427not comparableBGS counts manganese ORE at gross weight; USGS counts CONTAINED manganese. Ore grades ~30-50% Mn, which is the whole ratio.
cementcement, finished0.0732not comparableRESOLVED - not a unit slip: both series are in tonnes. It is a coverage hole. BGS cement has ~33 reporters led by Turkey, Germany, Poland, Italy and Spain; CHINA AND INDIA ARE ABSENT, and China alone is ~2.4 of the ~4.2 billion tonne world total. The BGS cement panel is a Europe-weighted sub-panel, not a world census, so its sum must never be used as a world denominator.
brominebromine1.609not comparableUnresolved: likely a compound-vs-element basis difference.
titaniumtitanium minerals1.4619not comparableBGS counts titanium MINERALS (ilmenite + rutile, gross); USGS reports on a TiO2-content basis.
bismuthbismuth, mine0.568not comparableUnresolved. Both sides are now MINE output (the USGS world mine series entered the cube on 16 Sep 2026; it ends in 2013, so only 2005-2013 overlap), and BGS national returns sum to about half of it. The earlier note blamed a stage mismatch - true only while the USGS side was refinery production.
boronborates1.2810not comparableBGS counts borate MINERALS (gross); USGS reports boron content.
graphitegraphite1.2716not comparableUnresolved: natural-graphite scope differs between the two.
micamica0.7920not comparableUnresolved: sheet vs scrap/flake mica scope differs.
gallium †gallium, primary0.815within 25%BGS gallium coverage is thin and USGS publishes an estimate rather than a census; neither side is a measured world total.
wollastonite †wollastonite1.197within 25%Within compilation tolerance; few producers, so one country revision moves the ratio noticeably.
seleniumselenium, refined1.1716within 25%Within compilation tolerance; refinery-output coverage differs by year.
germanium †germanium metal0.833within 25%BGS carries only 3-4 reporting countries for germanium metal, so its sum is not a world total; USGS states outright that reliable world estimates cannot be made. The near-agreement here is coincidence, not corroboration - this is the material that prompted the whole check.
feldsparfeldspar1.1552within 25%Within compilation tolerance. No structural basis difference identified; the two bodies revise on different schedules and BGS country coverage varies by year.
iron oreiron ore0.8746within 25%Iron ore scope differs: crude-ore vs usable-ore reporting is not consistent between the two bodies.
talctalc0.8736within 25%Within compilation tolerance; talc/pyrophyllite scope differs slightly.
magnesiummagnesite1.1023within 10%—
diatomitediatomite0.9227within 10%—
indiumindium, refinery0.9311within 10%—
rare earths †rare earth oxides1.076within 10%—
iodineiodine1.079within 10%—
perliteperlite1.0619within 10%—
tellurium †tellurium, refined0.946within 10%—
zirconiumzirconium minerals0.9516within 10%—
cobaltcobalt, mine1.0519within 10%—
saltsalt0.95101within 10%—
barytebarytes1.0435within 10%—
lithium †lithium minerals0.967within 10%—
arsenic †arsenic, white1.037within 10%—
gypsumgypsum and plaster1.0384within 10%—
tintin, mine1.0324within 10%—
asbestos †asbestos0.975within 10%—
rhenium †rhenium0.977within 10%—
vanadium †vanadium, mine1.024within 10%—
tungstentungsten, mine1.0220within 10%—
strontium †strontium minerals1.026within 10%—
nickelnickel, mine0.9827within 10%—
goldgold, mine1.0198within 10%—
fluorsparfluorspar0.9924within 10%—
mercurymercury1.019within 10%—
antimonyantimony, mine1.0117within 10%—
bauxitebauxite0.9931within 10%—
platinumplatinum group metals, mine0.9911within 10%—
phosphatephosphate rock1.0136within 10%—
leadlead, mine1.0146within 10%—
potashpotash1.0112within 10%—
silversilver, mine1.0168within 10%—
zinczinc, mine1.0151within 10%—
molybdenummolybdenum, mine1.0015within 10%—
cadmiumcadmium1.0018within 10%—
coppercopper, mine1.0054within 10%—
Withdrawn.
  • cobalt (2026-09-16): Not comparable (ratio 0.60), "resolved" as DR Congo artisanal output missing from BGS, with USGS world output said to climb to 294 kt in 2020. The USGS side summed two world series: mine production and refinery production, each 147 kt in 2020. Compared with mine production alone, the BGS sum agrees within 10% (ratio 1.05). The DR Congo explanation was built on the doubled figure and is withdrawn.

What this page does not claim

  • Not verification. Two compilations sharing the same national returns cannot verify each other.
  • Not accuracy, and not ground truth. Neither body measures the ore; both collect what is reported.
  • Agreement within 10% does not make a figure correct, and the band is wide.
  • A material that does not pair is not thereby worse — usually the two are simply counting different objects, such as gross ore against contained metal.
  • Where the two differ, this page does not declare the atlas’s number the winner.

How far the other agency sits from these figures. The numbers on this page are BGS. For the commodities where the atlas has measured it, the USGS world total for the same commodity and stage differs from the BGS one by a median of: graphite 30.7%, antimony 13.1%, cobalt 12.7%, tungsten 6.9%, rare earths 5.7%, copper 0.6% (median absolute difference over 2002–2024, unsigned). That is a disagreement between two compilations, not an error bar and not a correction of either — it can be measurement, definition, coverage or vintage. The same measurement also shows how far the USGS moves its own current-year estimate, which is the figure to check before treating a how firm is a current-year production figure?

figure?

Method. Pairings are declared explicitly in build_pairing.py, never inferred by matching material names: a ratio between “manganese ore, gross weight” and “manganese, contained metal” is a category error, not a disagreement. Both series are read from the atlas’s harmonised cube, which keeps every source’s native codes and units alongside the common labels. Data: BGS World Mineral Statistics and USGS Historical Statistics → pairing.json.