Critical Materials Atlas
Method · satellite · the commodity question

Which mineral is that mine?

The satellite page maps where mining scars the earth — but the polygons are all-commodity: they can’t tell lithium from coal. So we asked the obvious next question and measured the answer: overlay the best open, peer-reviewed, georeferenced mine database onto the footprint and see how much of it can actually be labelled. The result is the honest case for why this atlas builds material-level geography from production statistics and trade, using imagery only as a physical cross-check.

How the attribution works, and every caveat

We overlay Jasansky et al. (2023, Scientific Data) — a peer-reviewed, georeferenced database of mine facilities, each carrying a primary_commodity — onto the Maus (2022) all-commodity polygons. For each facility we test point-in-polygon; failing that, we attach the nearest polygon within 5 km. Each polygon is assigned to at most one commodity (inside beats near; nearer wins ties). Attributed footprint is then summed by commodity and mapped to the atlas’s 32 critical materials.

Caveats (stated in full): Jasansky covers 2,413 large, company-reported mines — not the artisanal/small sites that make up much of the Maus polygon count, so low coverage is expected and is part of the finding. We attribute footprint area by primary_commodity, the mine’s main product; the database also carries a free-text commodities_products field naming secondary products — we report those separately below but do not credit area to them, because a copper mine’s footprint is copper’s even when it yields trace cobalt. Generic “Coal” is not split into coking vs thermal, so it is not counted as the atlas’s coking coal. The 17% / 4% headline is a tier-2-buffer estimate, not a hard bound — see the sensitivity band below (inside-only to 25 km). Distances use nearest-centroid on WGS84 degrees (screening-grade, not projected). Temporal offset: Maus imagery ~2019, Jasansky compiled ~2021–23; the two share an author (V. Maus), so this is a consistency overlay, not a fully independent check. Sources: Maus et al. 2022 (PANGAEA) · Jasansky et al. 2023 (Zenodo, CC-BY) → commodity_attribution.json.

What the labelled footprint actually is

Attributed mine area (km²) by the facility database’s commodity classes. Green = maps to one of the atlas’s 32 critical materials; grey = not tracked (coal, gold, iron, silver, zinc, “other”).

The 32 critical materials: what open mine data can and can’t see

Of the atlas’s 32 tracked materials, only these are resolvable as a distinct commodity with mapped footprint in the open database:

Materialfootprint km²polygonswhere (top countries)

“But those mines yield lithium and cobalt too” — as byproducts, yes

A fair objection: the same database has a free-text commodities_products field that mentions more minerals. It does — but only as secondary products of a handful of large mines, and crediting a mine’s whole footprint to a trace byproduct would double-count. Here is every mention, coordinates or not, so you can judge:

Critical materialmines mentioning itwith coordinatesas

Tungsten, graphite, vanadium, beryllium and others don’t appear at all — not even in the free text. So the richer field doesn’t rescue a lithium or rare-earth footprint; it confirms these minerals surface, at most, as bylines in copper/nickel/gold operations.

Stacking every public register — and cross-checking them

The obvious question — would more mine registers help? — tested directly. We stack thirteen independent public sources onto Jasansky and re-run the join: USGS MRDS ( sites), OpenStreetMap ( commodity-tagged mines), the USGS critical-minerals deposit set ( curated points), a national cadastre (Geoscience Australia, ), and — crucially — the one dataset that covers artisanal mining, IPIS ( eastern-DRC & CAR sites), Wikidata ( mines, CC0), the USGS global mineral-operations file (reaching Russia/China/Indonesia), national geological surveys ( occurrences — Canada BC-MINFILE, Brazil SIGMINE, Finland’s GTK Fennoscandian database covering FI/SE/NO/NW-Russia, and Peru’s INGEMMET inventory), and the newest release — the ICMM Global Mining Dataset (2025) ( mine-type facilities, 47 commodities). Coverage triples — and because the sources are independent, where two label the same mine we can check whether they agree.

Sources joined to the satellite footprintfootprint labelledties to a critical material

Where the unlabelled half actually is

If more registers help, which ones? We mapped the still-unlabelled footprint by country. It is not scattered — it is concentrated, and mostly in countries that do not publish open, machine-readable mine data.

Countryunlabelled footprintopen mine data?

But is that unlabelled footprint hidden critical mines — or just coal and gravel?

The concentration above could mean two very different things: a wall of undisclosed lithium and rare-earth mines, or simply that the satellite sees enormous coal basins, clay pits and sand-and-gravel quarries that no critical-mineral register would ever list. We can tell them apart. For each of these countries we census every commodity-tagged mining point OpenStreetMap records and bucket it. This is a descriptive prior on the surrounding extraction, not a footprint measurement — but it bounds how much of the gap could plausibly be critical.

Countrytagged pointswhat the mining actually iscritical

How this compares to the published state-of-the-art

A fair question for any result: has someone done it better? Yes — and it is worth saying so plainly. The most advanced effort is Maus et al. 2026, the ERC “Mine the Gap” project (open data on Zenodo), which attributes a commodity to ~73% of a larger 145,000 km² mine-land footprint (26.8% unassigned) using a purpose-built clustering method, expert-validated at 95%.

Where we land against it. Our transparent stack of 13 public registers independently reaches 63% attribution on the Maus 2022 footprint — roughly 10 points behind the frontier, using nothing but open, individually-downloadable sources and a spatial join anyone can re-run (their clustering is more sophisticated, and their footprint base is larger). More telling is that the two agree on the shape, and we can check it directly against their published commodity shares:
Share of labelled footprintThis atlasMine the Gap
Coal43%22.5%
Gold13%21.1%
Copper18%6.6%
Iron13%
Coal is the single largest labelled commodity in both — the mine footprint is dominated by non-critical bulk, exactly the point of this page. The differences (our higher coal, their higher gold) trace to their larger footprint base capturing more small-scale artisanal gold, which our registers miss and which lands in our unlabelled remainder. A direct polygon-by-polygon overlap of the two isn’t possible yet — their per-cluster commodity vector isn’t openly released as of this writing (the January 2026 paper’s Zenodo record carries no downloadable file) — so this is a shares comparison, not a join. Where their bespoke method pushes coverage further, this page adds the piece they don’t: an explicit census of what the unlabelled footprint is made of. Independently reproducing most of the frontier’s coverage from fully open data, agreeing on the composition, and being candid about the gap is the honest claim here — and, as the next section shows, applying their method closes the rest.

Closing the gap with the frontier’s own method — district clustering

The frontier reaches 73% not with more data but with a better unit of analysis: it groups neighbouring polygons into a mining district and labels the district as a whole. A working open pit and its adjacent tailings dam, waste dumps and settling ponds are one operation mining one commodity — but our polygon-by-polygon join only labels the polygon that happens to sit near a registry point, leaving its own tailings pond blank. So we replicate the idea: union polygons whose centroids fall within a few kilometres, then propagate the district’s commodity to its unlabelled members. These propagated labels are a separate, lower-confidence tier — a spatial inference, not a direct source match — so we report the whole sensitivity band, not one number.

District scaleattributedcriticalpolygons propagated

Why the atlas reads production and trade, not imagery

Satellite polygons prove where the earth is disturbed, and the footprint page uses them exactly that way — as a physical cross-check on the producer story. But this page shows their ceiling for identity: even with the two largest public mine registers stacked on (above), well over half the mapped footprint stays unlabelled, and the minerals at the centre of every supply-risk debate — lithium, cobalt, rare earths, tantalum, tungsten — barely register as primary products in open mine data. Part of that 4% is an artifact of a large-mine registry meeting an all-commodity footprint; but the deeper limit is structural — the open taxonomy resolves ~11 broad classes and won’t name the criticals no matter how the buffer is tuned. So material identity has to come from sources that do resolve all 32 minerals: USGS & IEA production shares for the geography, reconciled bilateral trade for the flows. Imagery corroborates the footprint; it doesn’t name the mineral. This page is the receipt for that design choice.