Why Satellite Mapping for Geology?

Before we talk about wavelengths or algorithms, we need to answer a more fundamental question: why are we doing this at all? Traditional geological mapping is painstaking, expensive, and physically dangerous in high-altitude terrains. Satellite remote sensing does not replace the geologist in the field — but it extends their reach dramatically.

The Problem: Terrain That Cannot Be Mapped on Foot

The Nidar Ophiolite Complex sits in the Trans-Himalayan belt of Ladakh at elevations exceeding 4,000 m. Access is seasonal, roads are unreliable, and the terrain is steep and glaciated. A single field campaign can cover at most a few transects across a region spanning hundreds of square kilometres. Important lithological contacts — the boundary between a dunite body and serpentinized peridotite, for instance — can sit on a cliff face that no geologist will ever stand on.

This is the fundamental problem that satellite remote sensing solves. A satellite passes over the same area every few days, collects spectral data wall-to-wall, and delivers that data for processing from a desktop. Regional geological patterns become visible in hours rather than seasons.

Figure 1.1 — Satellite view of the Nidar Ophiolite Complex (Google Earth). High-altitude, structurally complex terrain like this is mapped continuously and synoptically from orbit, while ground traverses cover only narrow accessible corridors. Source: Ghaste (2020).

Figure 1.2 — Dunite outcrops along the Nidar Valley. The valley exposes the ultramafic mantle sequence (dunite, peridotite) and overlying mafic units in vertical succession — geologically ideal for remote-sensing validation but logistically challenging to traverse comprehensively on foot. Source: Ghaste (2020).

Traditional Mapping vs Satellite-Assisted Mapping

The Decision Framework

Before starting any satellite-based geological mapping project, ask these three questions:

Understanding Spectral Signatures

Before you can interpret a satellite image geologically, you need to understand why rocks look different at different wavelengths. This is not intuitive — two rocks that look identical in a photograph can be completely different spectrally in the shortwave infrared. The physics behind this is where geological remote sensing begins.

The Physics of Spectral Reflectance

When sunlight hits a rock surface, three things can happen to the incoming photons: they can be reflected (bounced back toward the sensor), absorbed (their energy taken up by the material), or transmitted (passed through, rare in opaque rocks). The reflectance at any given wavelength is simply the fraction of incoming energy that is reflected.

A reflectance spectrum plots this fraction against wavelength across the electromagnetic spectrum — from visible light (~400–700 nm) through near-infrared (NIR, ~700–1000 nm), shortwave infrared (SWIR, ~1000–2500 nm), and thermal infrared (TIR, ~8,000–12,000 nm). Geologically, the most diagnostic information lives in the SWIR, where molecular bonds in minerals produce characteristic absorption features.

The Three Key Absorption Mechanisms

Rock-Specific Spectral Fingerprints (from Nidar field samples)

The following table summarises the key absorption positions measured from field-collected specimens in the Nidar study and their physical causes. These spectral fingerprints are what classifiers use to separate rock types.

Figure 2.2a — ASD field-measured spectra (top) and ASTER-resampled spectra (bottom) for dunite sample Dun-1. The diagnostic Mg-OH absorption near 2250 nm — caused by serpentinization of olivine — is preserved when the high-resolution field spectrum is resampled to ASTER's 9 SWIR bands. Source: Ghaste (2020).

Figure 2.2b — ASD and ASTER-resampled spectra for gabbro (sample Gb-1). The characteristic W-shape double-trough between 2150–2350 nm is caused by pyroxene Fe absorption — a fundamentally different spectral fingerprint from dunite (Fig. 2.2a) that enables unambiguous lithological discrimination from satellite data. Source: Ghaste (2020).

How to Read a Spectrum: A Step-by-Step Sequence

Satellite Datasets Explained

Not all satellites see geology the same way. The choice of sensor directly controls which spectral features you can detect, how large a geological unit must be to appear in a pixel, and how much preprocessing is required. This section introduces the three sensors used in the Nidar study and teaches you to select the right tool for each geological objective.

Sentinel-2: High Resolution Context Mapping

Sentinel-2 is part of ESA's Copernicus programme and provides the highest spatial resolution of the three sensors used in this study. It carries the MultiSpectral Instrument (MSI) with 13 spectral bands. For geology, the most useful are the four 10 m VNIR bands (Blue, Green, Red, NIR) and the 20 m SWIR bands (Band 11 at 1610 nm and Band 12 at 2190 nm).

The limitation for geological work is critical: Sentinel-2 has only two SWIR bands, separated by a wide spectral gap. This means it cannot resolve the detailed absorption shapes that distinguish dunite from peridotite or detect the Mg-OH feature at 2300 nm with confidence. What Sentinel-2 is excellent for is providing spatial context — a high-resolution base map for tracing structural features and broad lithological units, particularly the contrast between mafic/ultramafic rocks and surrounding sedimentary sequences.

In the Nidar study, the Sentinel-2 FCC (Band 11 Red, Band 12 Green, Band 8A Blue) clearly distinguished mafic–ultramafic rocks (pinkish-brown tones) from the sedimentary suite (yellowish-green tones). However, within the ophiolite itself, Gabbro, Peridotite, and Indus Formation could not be separated because their spectral trends are too similar in the limited SWIR bands. Only Dunite, with its characteristic dip at ~2250 nm falling within Band 12, showed a distinctive signature.

ASTER: The Workhorse of Geological Mapping

ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) is the most powerful of the three sensors for geological mapping. It provides 14 spectral bands spanning VNIR (4 bands at 15 m), SWIR (6 bands at 30 m), and TIR (5 bands at 90 m). This combination is uniquely suited to ophiolite mapping because the SWIR bands capture mineral absorption features between 1.6–2.43 μm — exactly where serpentine, pyroxene, and carbonate produce diagnostic signals — and the TIR bands capture spectral emissivity variations tied to bulk silica content and mineral structure.

The key insight about ASTER's TIR capability: ultramafic rocks (peridotite, dunite) have high silica deficiency relative to mafic rocks (gabbro), and this difference is encoded in emissivity. Ultramafic rocks show higher emissivity at ASTER bands 10–12 and lower emissivity at bands 13–14, while gabbro shows the reverse pattern. This makes thermal emissivity a complementary and independent diagnostic tool.

Hyperion: Continuous Hyperspectral Coverage

Hyperion, aboard NASA's EO-1 satellite, is a hyperspectral imager providing 242 unique spectral channels from 450 nm to 2500 nm at 30 m spatial resolution and 10 nm spectral bandwidth. While Sentinel-2 samples the SWIR at only 2 points and ASTER at 6, Hyperion samples it at over 170 channels — enough to resolve the full shape of every absorption feature.

This spectral density enables a qualitatively different kind of analysis. Rather than detecting that an absorption exists somewhere near 2300 nm, Hyperion lets you resolve the exact shape and position of the Mg-OH feature, discriminating between different serpentine polymorphs (antigorite vs chrysotile vs lizardite) or quantifying the degree of serpentinization. The trade-off is a narrow swath width (7.75 km) and a more demanding preprocessing and noise-removal workflow.

Side-by-Side Comparison

Preprocessing Pipeline: From Radiance to Reflectance

A raw satellite image does not record the spectral properties of rocks. It records radiance at the sensor — electromagnetic energy that has been reflected from the surface, scattered by the atmosphere, and detected by the sensor. Before any geological analysis, this radiance must be converted into at-surface reflectance. This step is not cosmetic; it is the difference between a spectrum that physically describes your rock and one that describes your atmosphere.

Why the Atmosphere Matters

Sunlight must travel twice through the atmosphere: once on the way down to the surface, and once on the way back up to the satellite. In that journey, atmospheric gases (particularly water vapour, carbon dioxide, and ozone), aerosols, and haze do two things:

• Absorption: Specific wavelength ranges are absorbed almost completely (e.g., ~1400 nm and ~1900 nm are dominated by water vapour absorption — making spectral features at those wavelengths unreliable).
• Scattering: Short wavelengths (blue/UV) are scattered more than long wavelengths (Rayleigh scattering), adding a diffuse glow that inflates reflectance estimates at short wavelengths.

Without correction, a geologist comparing a SWIR spectrum from a satellite image to a laboratory ASD measurement will find a mismatch. The absorption features may be present, but their depth and shape will be systematically distorted by atmospheric effects. Classification based on uncorrected imagery will produce spectral mismatches — and ultimately, misclassified lithology.

FLAASH in Depth

FLAASH was applied in the Nidar study for both ASTER and Hyperion. For ASTER, suitable atmospheric parameters are selected via MODTRAN (Moderate Resolution Atmospheric Transmittance and Radiance), a physics-based radiative transfer model. A scaled DISORT MODTRAN multi-scatter model with 8 streams was used — this setting balances computational speed with accuracy for high-altitude arid terrain.

The effect of FLAASH is profound at key wavelengths. In the SWIR, where absorption features are diagnostic of mineral composition, uncorrected imagery may show artificially shallow or shifted troughs. After FLAASH correction, spectral profiles extracted from the image can be directly compared to field-measured ASD spectra and USGS library spectra.

Figure 4.2 — ASD field spectra (top) and ASTER-resampled spectra (bottom) for a peridotite sample. The characteristic Mg-OH absorption near 2300 nm — diagnostic of serpentinization — is clearly visible in both representations. After atmospheric correction, this feature is preserved in the satellite signal; without correction, it is masked by atmospheric water-vapour absorption. Source: Ghaste (2020).

Figure 4.3 — Complete ASTER image-processing workflow for lithological mapping within an ophiolite suite. After atmospheric correction, the SWIR branch (band ratios, spectral indices) and the TIR branch (emissivity, mineral indices) operate in parallel, producing complementary evidence streams that are fused at the final mapping stage. Source: Ghaste (2020), thesis Fig. 8.

Preprocessing Quality Checklist

Traditional Mapping Workflow

Once your imagery is atmospherically corrected, you have a stack of physically meaningful reflectance values. Now the geological interpretation begins. The traditional workflow follows a logical sequence: visualise, inspect, compress, index, classify, validate. Each step builds evidence.

A) False Color Composites (FCC)

A False Color Composite assigns different spectral bands to the red, green, and blue display channels. This exploits the human eye's sensitivity to colour to reveal spectral contrasts that are invisible in natural-colour images. In geological remote sensing, FCC is always the first step — before indices, before classification. It tells you whether there is any spectral structure worth analysing further.

Why ASTER 7-4-2 (Red-Green-Blue) for ophiolite mapping?

Each band is selected for a physical reason:

• Band 7 (Red, 2.266 μm) — Sensitive to Mg-OH-rich minerals and carbonates in serpentinized peridotites. Highlights zones with strong Mg-OH absorption.
• Band 4 (Green, 1.656 μm) — Shows high reflectance in dunite. Also highlights hydroxyl-bearing and serpentine minerals in serpentinized harzburgites/peridotites.
• Band 2 (Blue, 0.661 μm) — Highlights weathered iron minerals from olivine/pyroxene-rich peridotites (Fe³⁺ transitions).

The resulting FCC dramatically enhances contrast. In the Nidar study: Dunite appears bright green (high Band 4 reflectance); Peridotite appears dark brown; Gabbro appears light greenish; Lian Molasse appears cream-white; Chert-Jasper sediments appear bright yellow; Tso Morari crystalline appears greenish-yellow; and the contact between crystalline basement and the ophiolite is clearly traceable.

Figure 5.1 — ASTER False-Colour Composite (Band 7 → Red, Band 4 → Green, Band 2 → Blue) showing strong colour contrast between lithological units at the Nidar Ophiolite Complex. Each unit's apparent colour arises directly from differences in mineralogy: dunite appears bright green, peridotite dark brown, gabbro light greenish, the Lian Molasse cream-white, and the Tso Morari crystalline greenish-yellow. Source: Ghaste (2020), thesis Fig. 32.

Key Insight

In a well-designed FCC, each colour corresponds to a physical spectral behaviour. Geological interpretation begins by asking: "What mineral property explains this colour?" not "What colour looks like the rock I expect?" The latter leads to confirmation bias; the former leads to discovery.

B) Spectral Profiles and ROI Analysis

Once the FCC reveals potential lithological zones, you extract spectral profiles from Regions of Interest (ROIs) within those zones. An ROI is a set of pixels that represents a known (or hypothesised) geological unit. The shape of the extracted spectrum is your primary interpretive tool.

The V-shape vs W-shape diagnostic (from Nidar study)

In ASTER SWIR box plots extracted from ROIs at Nidar (Figure 5.2), a crucial diagnostic pattern emerges in the 2150–2400 nm window:

• Peridotite shows a characteristic V-shaped dip — a single broad trough centred near ~2300 nm, consistent with Mg-OH in serpentine minerals replacing olivine.
• Gabbro shows a W-shaped double dip — a reflection peak near 2250 nm (Band 7) flanked by two troughs at ~2200 nm and ~2350 nm, caused by pyroxene absorptions.
• Dunite shows elevated reflectance at Band 4 (1656 nm) compared to peridotite, then a dip at ~2250 nm from serpentine.

Figure 5.2 — Box plots of ASTER spectral profiles extracted from regions of interest (ROIs) for each lithology. The V-shape trough in peridotite between Bands 6–8 (2150–2350 nm) indicates Mg-OH absorption from serpentinization. Gabbro shows a W-shape trough reflecting pyroxene composition. Dunite shows elevated Band 4 reflectance. Shape-based discrimination is more robust than any single-band threshold. Source: Ghaste (2020), thesis Fig. 33.

D) Band Ratios and Spectral Indices

Band ratios are one of the most powerful tools in geological remote sensing. By dividing one band by another, you cancel out illumination effects (both bands are affected equally by brightness variations) and emphasise relative spectral features — the absorption contrasts that encode mineralogy. A ratio is high where the numerator band has high reflectance relative to the denominator; it is low where the numerator is absorbed.

Mafic Index (ASTER TIR Ratio 12/13)

The mafic index is derived from ASTER's thermal bands and exploits a fundamental property of silicate minerals: the position of the main emissivity trough (called the reststrahlen feature) shifts to longer wavelengths as silica content decreases. Ultramafic rocks (peridotite, dunite) have very low SiO₂ content, so their emissivity trough falls at shorter TIR wavelengths (bands 10–12). Mafic rocks (gabbro) have higher SiO₂, so their trough falls at longer wavelengths (bands 13–14).

Mafic Index Band 12 (9.075 μm) / Band 13 (10.657 μm)

High values (Band 12 brighter than Band 13) indicate mafic-to-ultramafic composition. In the Nidar scene, the mafic index image shows ultramafic units in bright white, with a sharp jump in value at the contact between the ultramafic unit and alluvium — an excellent geological boundary indicator.

Serpentine Index (ASTER SWIR Band Ratios)

The serpentine index maps the spatial extent of serpentinization — the alteration of olivine-rich ultramafics by hydrothermal water. It exploits the spectral behaviour of serpentine minerals (brucite, antigorite, chrysotile, lizardite), which show:

• A reflectance peak close to ASTER Band 5 (2.167 μm)
• A sharp dip at ASTER Band 7 (2.262 μm) due to the Mg-OH absorption feature

Serpentine Index (3-band FCC) Red = Band 3/Band 1  |  Green = Band 5/Band 7  |  Blue = Band 5/Band 3

This RGB combination displays serpentinized peridotites and dunites in pale green to yellow colours. The Band 5/Band 7 ratio is the key diagnostic: it is high (bright green) where Band 5 is elevated and Band 7 is depressed — exactly the spectral signature of serpentine minerals.

Figure 5.4a — ASTER mafic index (top panel) showing ultramafic units in bright white. The A–B transect profile (bottom panel) shows the sharp value jump across the lithological contact, demonstrating the index's ability to detect mineralogical boundaries. Source: Ghaste (2020), thesis Fig. 42.

Figure 5.4b — ASTER serpentine index — false-colour composite of bands 3/1, 5/7, 5/3 (RGB). Serpentinised zones appear in yellow-green tones, revealing the internal pattern of serpentinisation within the ultramafic unit. Where the mafic index identifies where the ultramafic rocks are, the serpentine index reveals how altered they are. Source: Ghaste (2020), thesis Fig. 43.

Key Insight: Physical Meaning First

Every index should be understood in terms of the mineral physics it encodes before it is applied. The serpentine index works because serpentine minerals have a specific peak-trough contrast in the 2150–2280 nm range. If you are working in a different geological setting (e.g., carbonates, phyllic alteration), different band combinations are needed. Never copy index formulas without understanding their physical basis.

Hyperspectral Analysis: The Power of Continuous Spectra

Everything in the traditional workflow improves with more spectral information. Hyperion provides hundreds of contiguous spectral channels, enabling a qualitatively different type of geological analysis — one that resembles the analysis a geochemist performs in a laboratory, now extended to the scale of a satellite scene.

Discrete vs Continuous Spectra: What Changes?

With Sentinel-2, you have 2 SWIR data points. With ASTER, you have 6. With Hyperion, you have over 170 contiguous samples across the same range. This matters because mineral absorption features have specific shapes — not just positions. A broad, asymmetric trough at 2290 nm tells a different story than a sharp, symmetric trough at the same position. With multispectral data, you can detect a feature is present; with hyperspectral data, you can characterise its shape.

Figure 6.1 — The same peridotite sample seen at two different spectral resolutions. The continuous ASD field spectrum (top) — analogous to a Hyperion observation — fully resolves the Mg-OH absorption feature near 2300 nm. The same spectrum resampled to ASTER's 9 SWIR bands (bottom) captures the feature only as a coarse two-point depression. Sentinel-2's broad SWIR bands (not shown) would not resolve it at all. Spectral resolution determines whether a sensor detects, approximates, or fully resolves a mineralogical absorption feature. Source: Ghaste (2020).

Hyperion Preprocessing: Noise First

Hyperion data requires more preprocessing than multispectral data. With 200+ bands, noise accumulates and many bands are unreliable (particularly around atmospheric absorption windows at ~1400 and ~1900 nm). The standard workflow applies:

1. Noisy band removal — bands dominated by water vapour absorption are excluded.
2. FLAASH atmospheric correction — as described in Section 4.
3. MNF (Minimum Noise Fraction) transform — separates signal from noise. MNF is a modified PCA that orders components by SNR rather than variance. The first MNF bands contain the most coherent signal; later ones are noise-dominated.
4. PPI (Pixel Purity Index) — identifies the most spectrally pure pixels (candidate endmembers) by iteratively projecting n-dimensional data and counting extreme pixels.
5. Endmember identification — inspect candidate endmember spectra in an n-D visualiser and match to known field spectra or library references.

Hyperion Band Indices Used in the Nidar Study

Three targeted indices were applied to the Hyperion image to map specific chemical signatures within the ultramafic terrain near Hanle:

High scores in the Mg-OH index indicate OH-altered and carbonate-bearing zones, normally concentrated in ultramafic rocks compared to volcano-sedimentary units. The ferrous index highlights zones of Fe²⁺-bearing mineralogy (olivine, pyroxene). Together, these three indices provide a chemical fingerprint of the alteration state within the ophiolite.

Figure 6.2 — Hyperion-derived mineral indices for the Hanle extension of the Nidar Ophiolite Complex. (a) Mg-OH Index — elevated values indicate serpentine-bearing ultramafics; (b) Carbonate Index — highlights hydrothermal carbonate veining associated with ocean-floor alteration; (c) Ferrous Index — distinguishes the iron-bearing mineralogy of fresh ultramafics from altered or oxidised facies. Source: Ghaste (2020), thesis Fig. 47.

AI-Based Geological Mapping: From Rules to Learning

Traditional remote sensing uses physics and expert knowledge to design features (ratios, indices) and decision rules (classifiers). Machine learning inverts this: instead of designing features manually, it learns them from labelled examples. This section explains how that transition works for geological mapping and where the physics-based and data-driven approaches complement each other.

Why Traditional Methods Have Limits

Classical methods like SAM and Maximum Likelihood work well when:

• Classes are spectrally distinct in the available bands.
• Training ROIs are clean and representative.
• The scene is spatially uniform enough that atmospheric correction is consistent.

They struggle when classes overlap spectrally (e.g., partially serpentinized peridotite vs fully serpentinized dunite), when spatial texture matters (e.g., banded ultramafics where the spatial pattern of alternating rock types is the signal), or when you want to simultaneously use many types of ancillary data (elevation, slope, curvature, published geological age). This is where machine learning adds value.

Feature Engineering: Building on Physics

The most powerful AI pipelines are not purely data-driven — they are physics-informed. Rather than feeding raw band values to a classifier, you construct a feature matrix that includes physically meaningful derived quantities:

Machine Learning Classifiers for Geological Mapping

Physics-Based vs Data-Driven: A Structural Comparison

Validation Protocol: Non-Negotiable Steps

Case Study: Nidar Ophiolite Complex, Ladakh

This section ties together every concept and method from the tutorial through a real geological example. The Nidar Ophiolite Complex (NOC) is an obducted slab of oceanic lithosphere exposed in the Trans-Himalayan terrain of Ladakh, NW India. Its lithological diversity — dunite, peridotite, gabbro, pillow lava, pyroxenite — makes it an ideal test bench for multi-sensor satellite mapping.

Geological Setting

Ophiolites are fragments of ancient oceanic crust and upper mantle that have been thrust onto continental margins during tectonic collisions. The Nidar Ophiolite records the closure of the Neo-Tethys Ocean as the Indian subcontinent collided with Asia. From bottom to top (south to north in the NOC), the exposed section follows the classic ophiolite stratigraphy:

Figure 8.1 — The Nidar Ophiolite Complex study area. The from-south-to-north section through the complex reveals the plutonic sequence (dunite, peridotite, gabbro) succeeded by pillow lavas and overlying sedimentary units. Field sample collection points guided the selection of training regions of interest (ROIs) for all classifiers in this study. Source: Ghaste (2020), thesis Fig. 9.

Multi-Sensor Evidence Chain

The Nidar study was not solved by a single dataset. Each sensor contributed something the others could not:

Figure 8.2 — ASTER Spectral Angle Mapper (SAM) classification of the Nidar Ophiolite Complex. (a) SAM classification for peridotite. (b) SAM classification for dunite. (c) SAM classification for gabbro. (d) SAM classification for pillow lava. Together with the mafic index (Fig. 5.4a), serpentine index (Fig. 5.4b), and published field geology, these classification products form a multi-line evidence stack — convergence between independent products gives confidence in the final lithological interpretation. Source: Ghaste (2020), thesis Fig. 38.

Key Geological Findings

• SAM classification on ASTER SWIR data successfully discriminated dunite, peridotite, gabbro, and pillow lava within the ophiolite suite.
• Peridotite and dunite are the most serpentinized units, validated by both spectral indices and thin section analysis showing mesh texture and serpentine veins around relict olivine.
• Gabbroic intrusions within the peridotite mantle unit are geologically expected and confirmed by image analysis — visible as isolated gabbro (red) pixels within the peridotite-dominated zone.
• Hyperion data extended the analysis eastward to Hanle, confirming peridotite presence far from the main Nidar Valley outcrops.
• ASTER is the most effective single sensor for regional lithological classification because of its unique combination of SWIR and TIR coverage.