Using Sentinel-2 and ASTER in the Mountains: A Field Note From Gilgit-Baltistan

Last March I was sitting at 3,800 meters near Skardu, laptop balanced on a rock, trying to figure out why a hydrothermal alteration zone I'd mapped from Sentinel-2 the night before had completely disappeared in the ASTER scene I pulled the next morning.

Turned out the ASTER scene was from 2007. Sentinel-2 was from last week. A 17-year gap. Vegetation had crept in, a small landslide had wiped out part of the outcrop, and the contrast I was excited about was partly seasonal snow melt.

I got this wrong at first. I used to treat satellite data like it was a single truth. Now I treat it like witness testimony — useful, but every witness has a bias and a date stamp.

This post is about how we actually combine Sentinel-2 and ASTER at GeoMine AI for exploration work in mountain terrain — the kind of steep, shadowed, snow-patched, structurally complex ground that makes up most of northern Pakistan.

Why these two sensors, and why together

Sentinel-2 gives you 13 bands, 10-meter resolution on the visible and near-infrared, and a 5-day revisit. It's free. It's recent. The SWIR bands (B11 and B12) at 20 meters are surprisingly useful for picking up iron oxides, clay groups, and broad alteration footprints.

ASTER is older — Terra launched in 1999 — but it has something Sentinel-2 doesn't. Five thermal infrared bands and six SWIR bands between 1.6 and 2.43 micrometers. That SWIR range is where Al-OH, Mg-OH, and carbonate minerals show their diagnostic absorption features. If you're hunting for argillic, phyllic, or propylitic alteration around a porphyry copper target, ASTER SWIR is still the workhorse despite the SWIR detector failing in 2008 (we mostly use pre-2008 scenes for SWIR work, post-2008 for VNIR and TIR).

So the logic is simple. Sentinel-2 for fresh, high-resolution context. ASTER for the spectral depth you can't get anywhere else for free.

Here's the thing — neither one alone is enough in mountains.

What the mountains do to your data

Karakoram and Hindu Kush terrain breaks satellite imagery in specific ways, and you need to know them before you trust any false-color anomaly map.

Shadow. Slopes steeper than 35 degrees throw shadows that swallow reflectance values. On a north-facing wall in winter, you're looking at noise, not geology. We mask anything below a solar incidence angle of about 15 degrees before running ratios.

Snow and ice. Snow has a very bright VNIR signature and a collapsing SWIR signature that can mimic clay alteration if you're sloppy. NDSI (Normalized Difference Snow Index) takes care of obvious snow, but firn and dirty ice on glacier margins are sneaky. One of my mines near Shigar sits next to a debris-covered glacier and I've seen consultants flag the moraine as a phyllic alteration zone. It's not. It's just wet rock flour.

Topographic correction. This is non-negotiable. We run a Minnaert or C-correction using SRTM 30m or ALOS PALSAR 12.5m DEM before any band ratio work. Skip this step in the mountains and you're mapping illumination, not mineralogy.

Atmospheric variation with elevation. A scene that covers ground from 1,500m to 6,000m has very different atmospheric path lengths. Sen2Cor handles Sentinel-2 reasonably well. For ASTER we use crosstalk correction plus a regional atmospheric model. It matters.

A workflow that actually produces drill targets

For a typical project — say a 400 sq km block in Chitral or Kohistan — here's roughly how it goes.

First, Sentinel-2 mineral exploration at the regional scale. We pull 3 to 5 scenes from different seasons, build a cloud-free composite, run topographic correction, then generate iron oxide ratios (B4/B2), ferric iron (B11/B8), and a broad clay index (B11/B12). This gives us first-pass alteration footprints at 20m resolution across the whole block in about a day.

Then we overlay ASTER spectral analysis on the zones that lit up. ASTER band ratios — (B4+B6)/B5 for Al-OH (kaolinite, alunite, muscovite), (B6+B9)/(B7+B8) for Mg-OH and carbonates, B4/B8 for argillic — narrow the footprint to specific mineral assemblages. A broad iron oxide zone in Sentinel-2 might break into a sericite core and a propylitic halo in ASTER. That's the kind of zoning a copper-gold geologist actually wants to see.

Then the AI layer. We train on known deposit signatures — the Reko Diq alteration spectrum, the Saindak halo, chromite-bearing ultramafics from Muslim Bagh — and ask the model where else in the scene those patterns repeat. It flags candidate zones, ranks them by similarity score, and we hand the top 20 to a field geologist.

Out of those 20, maybe 6 to 8 are worth a boots-on-the-ground visit. Of those, 2 or 3 become real drill targets. Last year on one block we ran this workflow on, the hit rate from satellite-flagged target to confirmed alteration on the ground was 38%. Not perfect. But compared to walking 400 sq km blind, it's a different sport.

Where it falls apart

Honestly, satellite mineral mapping doesn't find deposits. It finds places worth looking. Anyone selling you a map that says "gold is here" from Sentinel-2 alone is selling you a story.

The sensors see the top few microns of surface rock. A blind deposit under 30 meters of cover is invisible to both Sentinel-2 and ASTER. That's where magnetics, gravity, and IP come in — and I've written about those separately.

Also, vegetation. Below about 2,500m in parts of Azad Kashmir and lower Chitral, oak and pine cover defeats SWIR-based alteration mapping pretty thoroughly. We can sometimes pick up geobotanical stress signals in Sentinel-2 red-edge bands, but it's a much weaker signal than bare-rock spectroscopy.

And dust. Pakistan's mountains in May and June are dusty. A thin loess coating on outcrop can mute the very absorption features you're trying to detect. We've started flagging scenes acquired within a week of major dust events and just not using them.

If you're running geomining work in the Karakoram or anywhere similar and you're only using one sensor, you're leaving information on the table. The combination — Sentinel-2 freshness plus ASTER spectral depth plus a proper DEM plus someone who's actually walked the ground — is what makes the difference between a pretty map and a drill hole that hits.

What's the alteration assemblage you're trying to map this season?