C5 · Publication Volume 15

Remote Sensing, Aerial Imagery and Spectral Geology

Builds geological interpretation of imagery, terrain and spectroscopy from sensor physics and product provenance.

Purpose and boundary of this book

Remote sensing measures energy from a surface or volume without placing the detector at the material being interpreted. An image pixel is not a rock sample and a colour is not a mineral name. The recorded value is produced by illumination or transmitted energy, surface and atmospheric interaction, sensor response, viewing geometry, spatial support, calibration, preprocessing and noise. Terrain, vegetation, regolith, water, cloud, shadow and human disturbance can alter or conceal the expression of geology. The central scientific task is therefore to connect a traceable measurement to a geological hypothesis without turning visual similarity into certainty.

This book develops that evidence chain for optical and terrain observations used in geological work. It covers the electromagnetic spectrum; passive and active sensors; spatial, spectral, radiometric and temporal resolution; image geometry; radiometric and atmospheric correction; co-registration; multispectral composites and indices; mineral reflectance spectroscopy; imaging spectroscopy; digital elevation models and lidar; cover effects; change detection; product identifiers, quality masks, licences and provenance. The emphasis is interpretation that can be reproduced, challenged and checked in the field.

The tutorial does not prescribe a preferred platform, instrument, product, band ratio, classifier, cloud threshold, spatial resolution or spectral library. Suitability depends on the question, scale, material exposure, wavelength sensitivity, atmosphere, illumination, terrain, acquisition date, calibration, quality flags and consequence of error. It does not provide legal, safety, environmental, procurement, investment, resource-estimation or operational approval. A spectral or image anomaly is a measured difference requiring explanation; it is not proof of a mineral, lithology, deposit, disturbance cause or economic outcome.

General and institution-neutral scope

This is a general, institution-neutral tutorial. It has no affiliation with, sponsorship by, endorsement from or curriculum relationship to any company or individual. It is not written for a named operator, contractor, consultancy, university, public agency, software product, sensor brand, licence, property, mine or private dataset. Every unnamed scene, flight line, spectrum, pixel, terrain model, target and locality in an example is synthetic teaching material.

People, public bodies, standards organisations, missions, journals and repositories named in source notes identify technical sources only. A citation does not make any person or organisation the author, publisher, sponsor, provider, partner, endorser, scientific authority or curriculum subject of this tutorial. The website that hosts these pages is only a delivery surface. It is not the tutorial's author, publisher, sponsor, provider, scientific authority or curriculum subject, and it asserts no ownership of the curriculum.

This separation is scientifically useful. A familiar logo cannot repair an unrecorded processing baseline, mixed pixel, cloud shadow, spectral smile, vertical-datum error or misregistered time series. Conversely, a modest public product can support a strong decision when its identity, physics, corrections, masks, uncertainty and field checks are explicit. Claims in this book stand or fall on traceable observations and reproducible reasoning.

The observation-to-decision chain

Use the following sequence throughout the book:

  1. State the geological question, decision and at least two viable explanations.
  2. Predict which surface or terrain properties would express each explanation, at which wavelengths, spatial scales and seasons.
  3. Select a sensor and product whose spectral response, footprint, dynamic range, revisit and geometry can test those predictions.
  4. Preserve the original scene, product identifier, processing level, acquisition time, orbit or flight geometry, calibration and licence.
  5. inspect quality masks, saturation, cloud, haze, shadow, water, snow, terrain and coverage before computing a geological derivative.
  6. correct geometry and radiometry only with declared algorithms, auxiliary inputs, parameters and output units.
  7. co-register observations at a precision adequate for the smallest change or boundary that matters.
  8. examine spectra, bands and terrain derivatives in physical units before classification or visual enhancement.
  9. test geological, atmospheric, biological, surficial, geometric, sensor and human explanations for each material feature.
  10. report observations, inference, confidence, excluded support, field checks and the result that would reverse the decision.

The chain must work in reverse. A reviewer should be able to start from a coloured map, identify the contributing pixels, recover masks and transformations, locate the source granules and reconstruct acquisition conditions. A screenshot, tile cache or exported classification without that route is not a scientific record.

Learning outcomes

After completing the book, a learner should be able to:

  • relate wavelength, frequency and photon energy to reflection, absorption, emission and active returns;
  • distinguish spatial sampling from resolving power, and spectral sampling from diagnostic feature resolution;
  • evaluate radiometric quantisation, signal-to-noise ratio, saturation and temporal sampling together;
  • audit orthorectification, atmospheric correction, topographic correction and co-registration;
  • choose band combinations and indices from a stated physical hypothesis rather than colour preference;
  • measure absorption position, depth, width, asymmetry and continuum while respecting mixtures and grain effects;
  • compare library, field and image spectra only after wavelength, response function, geometry and scale are reconciled;
  • build a hyperspectral mapping workflow with bad-band masks, dimensionality control and confidence estimates;
  • derive and review elevation, slope, aspect and hillshade without treating illumination artefacts as structure;
  • separate regolith, vegetation, moisture, cloud, shadow and human disturbance from plausible geological expression;
  • design change detection that controls registration, seasonality, sensor differences and missing observations; and
  • release a provenance-aware interpretation with scene identifiers, processing levels, masks, licences, uncertainty and field tests.

Prerequisites and notation

The concept map places this book after A1 quantitative foundations, A3 Earth materials, B4 geomorphology and regolith, and C2 target generation. C1, C3 and C4 are useful when mineral-system, geochemical and geophysical evidence are integrated. Learners should be comfortable with units, functions, logarithms, vectors, matrices, coordinates, raster grids, probability and competing hypotheses.

Wavelength is \lambda, frequency is \nu, and the speed of light is c=\lambda\nu. Spectral radiance is L_\lambda and irradiance is E_\lambda. A simple directional reflectance factor may be written \rho_\lambda=\pi L_\lambda/E_\lambda only when its illumination and viewing assumptions are stated. A sensor band value depends on its spectral response S_i(\lambda), approximately d_i=\int L_\lambda S_i(\lambda)\,d\lambda after calibration terms. These symbols do not imply that every delivered product uses the same convention.

Spatial coordinates require a declared coordinate reference system and pixel convention. Elevation needs a horizontal and vertical datum. Time uses a declared time standard. Reflectance is dimensionless; radiance has radiometric units; digital number is a stored code, not a physical unit; elevation, slope and aspect must declare units and reference. Never infer a scale factor, no-data value, wavelength centre, grid origin or acquisition time from display appearance.

How to use the diagrams and synthetic cases

Each numbered lesson contains one purpose-built diagram. Spectral curves are schematic and show reasoning relationships rather than reference spectra. Image grids illustrate support, mixing, offsets and masks; they are not maps of real places. Terrain shading demonstrates how a derivative changes with parameters and does not prescribe a visual style. Arrows indicate dependency or transformation, not institutional workflow.

All unreferenced numbers and locations are synthetic. They expose units, calculation, sampling and failure modes; they are not sensor specifications, regulatory thresholds, mineral abundances, expected accuracies or success rates. A worked answer must include the hypothesis, product identity, units, masks, calculation, sensitivity to uncertainty, alternative explanation and decision implication. Matching a colour or a library curve without those elements is incomplete.

Reproducible remote-sensing records

Treat source granules and their metadata as immutable. Calibration, masking, reprojection, resampling, mosaicking, compositing, continuum removal, dimensionality reduction, classification, terrain derivation and change detection create versioned outputs rather than replacements. A reproducible package links collection and granule identifiers; version and processing baseline; acquisition and production times; platform and sensor; wavelength and response metadata; spatial grid and datums; radiometric scale and offset; no-data and saturation codes; quality masks; auxiliary atmosphere and terrain inputs; algorithm and parameters; software-independent equations; random seeds where relevant; training and validation support; uncertainty; licence; citation; and released products.

Every derived raster needs a machine-readable transformation record. Preserve masks separately from modified values. If pixels are removed, state the rule and count. If several scenes are composited, retain the contributing scene and observation date per pixel where practical. If exact locations are restricted, publish an authorised generalisation without breaking the private lineage. A rendered image is an inspection view, not the underlying measurement.

Assessment and completion standard

Completion requires a synthetic, platform-neutral interpretation package containing:

  1. a decision and at least two geological or non-geological hypotheses;
  2. predicted wavelengths, spatial scales, temporal behaviour and confounders;
  3. a product-selection matrix covering all four resolution types and processing level;
  4. an immutable source inventory with identifiers, versions, licences and checksums;
  5. geometry, radiometry, atmosphere, terrain and co-registration checks;
  6. explicit cloud, shadow, saturation, water, vegetation and disturbance masks;
  7. multispectral or spectroscopic measurements in physical units with uncertainty;
  8. a terrain or time-series analysis whose parameter sensitivity is shown;
  9. field or independent validation designed to distinguish the remaining hypotheses; and
  10. a conclusion separating observation, inference, confidence, excluded support and reversal conditions.

The package passes when another reviewer can reproduce principal outputs, trace every classified pixel to source observations, explain every correction and mask, identify unsupported areas and state which new observation could change the action. A visually convincing map without product identity, masks, provenance and alternatives does not meet the standard.

Core sources