Rare Earth Elements Exploration Using Multispectral Data: Honest Notes From the Field

Last March, I was standing at 3,200 meters in Skardu, looking at a rust-colored outcrop my foreman swore was "something special." Turned out he was half right. The XRF showed elevated thorium and cerium — classic REE-bearing carbonatite signal. But here's the part that bothers me: we'd already flagged that exact pixel six weeks earlier from a desk in Islamabad, using free Sentinel-2 data and a bit of band-ratio work.

That's the story of rare earth element exploration in 2026. The satellites have been screaming at us. We just haven't been listening properly.

Why REEs Are Harder to Map Than Copper or Gold

Rare earth elements don't behave like base metals. You can't just look for an oxidation halo and call it a day. REEs — neodymium, dysprosium, cerium, lanthanum, the whole lanthanide family — usually sit inside host minerals like monazite, bastnäsite, and xenotime. These minerals have weird, narrow absorption features in the visible and near-infrared range, mostly driven by the f-orbital electron transitions of the REE ions themselves.

Neodymium has sharp absorption near 580nm, 740nm, and 800nm. Dysprosium shows up around 910nm and 1100nm. Samarium has a feature near 1550nm.

Problem? Sentinel-2's bands are 20 to 60nm wide. ASTER's SWIR bands are wider still. So you're trying to catch a 10nm-wide absorption feature with a 40nm-wide net. Most of the signal gets washed out.

And yet. We still find targets. Because REE deposits rarely show up alone — they come with friends.

What Multispectral Actually Catches (and What It Misses)

Here's what I tell new geologists who join geomines: stop hunting the REE directly. Hunt the host rock and the alteration.

Most economic REE deposits in Pakistan are tied to one of three settings — carbonatites (think Sillai Patti in Khyber Pakhtunkhwa), alkaline intrusions, and ion-adsorption clays weathered from granites. Each leaves a multispectral fingerprint we can actually see.

Carbonatites produce a calcite + dolomite + apatite signature. ASTER bands 6, 7, 8 pick up the carbonate absorption around 2.31–2.33 micrometers beautifully. Run the (B6+B8)/B7 ratio and bright pixels are your friends. We've used this across the Peshawar Plain alkaline complex and gotten field-confirmed hits about 4 out of 11 times — not amazing, but for free satellite data, I'll take it.

Alkaline intrusions often weather to produce iron oxide staining plus clay alteration. Sentinel-2 B4/B2 for iron, B11/B12 for clay. Stack them, mask the vegetation with NDVI under 0.2, and the targets light up.

Ion-adsorption clay deposits — the kind China built its REE empire on — are trickier. They look like normal weathered granite from space. But thorium often travels with the light REEs, and radiometric data (when we can get it) plus subtle kaolinite signatures in B11 and B12 give us something to chew on.

What multispectral won't do? Tell you the grade. Tell you which REE. Tell you if it's cerium-dominant (cheap, abundant) or dysprosium-heavy (the stuff EV magnets need, trading at roughly $387/kg last I checked). For that, you drill. Or you fly hyperspectral, which costs about 80x more per square kilometer.

I used to think we could push multispectral further than this. Spent eight months in 2023 trying to fit narrow absorption models to Sentinel-2 data. Honestly? It mostly didn't work. The spectral resolution isn't there. What works is using multispectral to narrow 10,000 square kilometers down to maybe 40 square kilometers worth of drill-worthy targets. Then you spend money on the ground.

How We're Doing It at GeoMine

Our REE remote sensing workflow at geomines runs roughly like this. Pull Sentinel-2 L2A scenes — cloud cover under 10%, dry season only (October to February for most of northern Pakistan). Stack with ASTER L1T where coverage exists. Add SRTM 30m DEM for slope and drainage analysis because REE-bearing pegmatites love structural intersections.

Then we run a stack of band ratios: carbonate index, ferric iron index, AlOH clay index, and a custom thorium-proxy we built from spectral mixing analysis on known deposits in Sillai Patti and Loe Shilman. The AI layer on top weighs these against lithology maps from GSP and known geochemistry from old reports — some of which date back to 1978 and were typed on actual typewriters.

Output is a heatmap. Red pixels mean go look. Yellow pixels mean maybe. Everything else, ignore.

Of the 15 mines I personally own in Gilgit-Baltistan, three were originally flagged this way before I ever set foot on them. Two turned out to be marble (not what we were after, but still valuable). One is now in early-stage REE evaluation with samples sitting at a lab in Lahore.

The Part Nobody Talks About

Pakistan has documented REE occurrences in at least 14 locations across KP, Gilgit-Baltistan, and Balochistan. The Sillai Patti carbonatite alone has estimated reserves that, if even halfway accurate, would put Pakistan in the top 20 globally for light REE potential.

But here's the thing — almost nobody is exploring these systematically. The big mining companies are still focused on copper at Reko Diq. The government's geological survey is underfunded. And the satellite data that could flag a hundred new targets in a single afternoon is sitting unused on ESA and USGS servers.

That gap is the entire reason geomining as a discipline exists. Cheap eyes from orbit. Smart algorithms in the middle. Boots on the ground only where the data says it's worth the diesel.

Will multispectral solve Pakistan's REE puzzle on its own? No. You still need geochemistry, you still need drilling, you still need someone willing to hike up a scree slope in July when it's 41 degrees and the helicopter is grounded.

But if you're an investor or a mine owner reading this and you haven't even pulled a free Sentinel-2 scene over your concession yet — what exactly are you waiting for?