From Satellite Image to Drill Hole: What Actually Happens Between Pixel and Pickaxe

Last March, I stared at a Sentinel-2 image of a ridge in Skardu for about 40 minutes before I noticed it. A faint yellow-orange smear, maybe 180 meters across, sitting on the hangingwall of a fault I'd missed twice. Six weeks later we had a drill rig there. Eight weeks after that, visible chalcopyrite in core.

That's the pipeline. But most people only see the start and the end — the pretty satellite map on one side, the drill assay on the other. The middle is where exploration actually happens, and it's where 90% of projects die.

So let me walk you through it the way I'd explain it to a friend who just bought a lease and has no idea what to do next.

Stage 1: Pixels, not maps

Everybody calls it "satellite imagery." That's misleading. What we're actually working with at GeoMine AI are stacks of spectral bands — Sentinel-2 gives us 13, ASTER gives us 14, and each one sees a different slice of light the human eye can't.

Iron oxides show up loudest in bands 4 and 11. Clay alteration (the kind that wraps porphyry copper systems) lives in ASTER bands 5 through 8. Silica caps — the smoking gun for epithermal gold — pop in the thermal infrared. None of this is visible in a Google Earth screenshot. You need the raw data.

First step is always atmospheric correction. Skip it and your "anomaly" might just be a cloud shadow or a wet streambed. I learned this the hard way on a chromite target in Muslim Bagh — spent two weeks excited about something that turned out to be a dust plume.

After correction, we run band ratios and principal component analysis. The output isn't a map yet. It's a probability surface. A heatmap of where the rocks are behaving weirdly.

Stage 2: Structure, because minerals don't float

Here's the thing most non-geologists miss. A spectral anomaly by itself means nothing. Minerals concentrate along structures — faults, shear zones, fold hinges, contact boundaries. No structure, no deposit. Doesn't matter how pretty your alteration map looks.

This is where SRTM DEM data earns its keep. We pull 30-meter elevation models, run hillshade from four directions, then trace lineaments. Long straight valleys. Abrupt ridge offsets. Drainage patterns that bend for no obvious reason. Those are your faults, even when no geologist has ever walked the ground.

Then we add SAR (Sentinel-1) for surface roughness and moisture. SAR sees through clouds, which matters in Gilgit Baltistan where half the year you can't get an optical pass worth using.

Overlay the spectral anomaly on the structural map. The places where strong alteration sits on or near a major lineament — those are your targets. Not the whole anomaly. Just the intersections. Usually you go from a 50 km² area of interest down to maybe 4 or 5 spots worth visiting.

This is the part our breeze geo mineral analysis engine automates. What used to take a senior geologist three weeks of GIS work now runs in about 90 minutes.

Stage 3: Ground truth, or you're gambling

No one drills off satellite data alone. Anyone who tells you otherwise is either lying or about to lose money.

Once we've ranked targets, somebody has to actually go there. On my own mines I do this myself — partly because I trust my own eyes, partly because helicopter time in GB is brutal on the budget. You're looking for:

• Outcrop that matches the predicted alteration (gossan, silicification, malachite staining, whatever the model said should be there)
• Structural confirmation — is the fault really there, what's its dip, what's the sense of movement
• Rock samples for assay and petrography

The sample results either confirm the satellite story or they don't. Honestly, about 1 in 3 of our high-confidence targets fall apart at this stage. That's not failure — that's the system working. Better to kill a target with a $400 assay than a $400,000 drill program.

For the ones that survive, we tighten the grid. Trenching. Soil geochem on a 50m by 25m pattern. Sometimes ground IP or magnetics if the budget allows. By the end of this stage you should have a 3D mental model of what you think is underground.

Stage 4: Drill collar placement

This is the moment the whole pipeline has been building toward. Where do you put the hole?

A bad collar wastes everything. I've seen companies drill straight down through a steeply dipping vein and miss it by 12 meters. I've seen holes collared on the wrong side of a fault, drilling away from the mineralization instead of into it.

The answer comes from combining everything — the spectral signature tells you what's there, the structural model tells you the geometry, the surface sampling tells you the grade direction, and the topography tells you what's actually drillable with the rig you can afford to get up the mountain.

We usually plan three holes minimum. One to test the core of the model, two to test the edges. If all three hit, you've got a deposit. If only the middle one hits, you've got a small lens. If none hit — back to Stage 2, figure out what your model got wrong.

What this actually costs

A traditional exploration program from regional reconnaissance to first drill hole runs $180k to $400k and takes 14-18 months. Using the satellite-first approach we run at geomines, we've gotten that down to about $45k and 5 months on my own properties in Shigar and Roundu.

Is the satellite approach perfect? No. We still miss things. Deeply buried deposits with no surface expression are basically invisible to us — you need geophysics for those, and that's a different conversation.

But for the kind of shallow, structurally controlled deposits that make up most of Pakistan's $6 trillion in untapped reserves? The pipeline works. I've drilled the holes. I've seen the core.

The question isn't whether satellite intelligence belongs in your exploration pipeline. It's why you're still running programs without it.