15.1 Principles of remote sensing in geomorphology

Remote sensing for geomorphology

Remote sensing allows geomorphologists to study Earth's surface without being physically present. By measuring electromagnetic radiation that's reflected or emitted from the ground, researchers can map landforms, track surface processes, and detect changes across scales ranging from individual boulders to entire mountain ranges.

Fundamentals of remote sensing

The core idea is straightforward: a sensor (on a satellite, aircraft, or drone) detects electromagnetic energy coming from Earth's surface, and that energy carries information about what's down there.

Different parts of the electromagnetic spectrum reveal different things:

• Visible light detects surface features, colors, and textures
• Infrared radiation measures surface temperature and vegetation health
• Microwave radiation penetrates clouds and vegetation canopy, making it useful for terrain mapping in any weather

Spectral signatures are what make identification possible. Every surface material (rock, soil, water, vegetation) reflects and emits radiation in a characteristic pattern across wavelengths. By analyzing these patterns, you can distinguish limestone from sandstone, healthy vegetation from stressed vegetation, or dry soil from saturated soil.

Spatial resolution determines the smallest feature you can actually see in an image:

• High resolution (< 1 m): Individual boulders, small gullies, building-scale features
• Medium resolution (10–30 m): River channels, glacial moraines, landslide scars
• Low resolution (> 250 m): Regional landform analysis like mountain ranges or coastal plains

Remote sensing parameters and systems

Beyond spatial resolution, two other resolution types matter:

• Temporal resolution is how often a sensor revisits the same area. Daily observations can track rapid events like landslides or volcanic eruptions, while monthly or yearly revisits suit slower processes like coastal erosion or glacier retreat.
• Radiometric resolution describes how finely a sensor distinguishes differences in energy. A 12-bit sensor records 4,096 brightness levels per band, while a 16-bit sensor records 65,536. Higher radiometric resolution helps detect subtle variations in surface moisture or mineral composition.

Remote sensing systems fall into two categories based on their energy source:

Remote sensing platforms and sensors

Satellite-based platforms

Earth observation satellites provide global coverage with regular repeat cycles:

• The Landsat series has been collecting multispectral imagery since 1972, giving researchers over 50 years of continuous data for change detection studies.
• Sentinel missions (part of the EU's Copernicus programme) provide free, open-access data for land monitoring, maritime surveillance, and atmospheric studies.

High-resolution commercial satellites push spatial detail further. WorldView-3 achieves 31 cm panchromatic resolution, fine enough to map individual rock outcrops. IKONOS was the first commercial satellite to offer sub-meter resolution imagery for civilian use.

Synthetic Aperture Radar (SAR) satellites operate in the microwave region. They penetrate cloud cover and work day or night, and they can measure surface deformation (subsidence, tectonic movement) with millimeter-scale accuracy using a technique called interferometry.

Airborne and ground-based platforms

Manned aircraft offer flexible data collection. Aerial photography provides high-resolution imagery for detailed mapping, and airborne LiDAR generates digital elevation models with 1–10 cm vertical accuracy, which is precise enough to detect subtle fault scarps or floodplain microtopography.

Unmanned Aerial Vehicles (UAVs/drones) have transformed small-area surveys. Structure-from-Motion (SfM) photogrammetry uses overlapping drone photos to build detailed 3D surface models without expensive LiDAR equipment. Thermal cameras mounted on drones can map temperature variations in geothermal areas or track water seepage.

Ground-based platforms fill in the gaps:

• Terrestrial LiDAR scanners capture extremely detailed 3D point clouds of cliff faces, outcrops, or active landslides, often at centimeter or sub-centimeter resolution.
• Ground-penetrating radar (GPR) sends radar pulses into the subsurface to image buried structures and stratigraphy, one of the few remote sensing tools that looks below the surface.

Specialized sensors for geomorphology

Multispectral sensors capture data in several discrete wavelength bands. Landsat 8's OLI sensor has 11 spectral bands spanning visible through thermal infrared. ASTER provides 14 bands, with strong thermal infrared coverage that's particularly useful for geological and mineral mapping.

Hyperspectral sensors take this much further, collecting data across hundreds of narrow, contiguous bands. The airborne AVIRIS sensor, for example, records 224 bands. This density of spectral information enables detailed analysis of surface mineralogy and soil properties that multispectral sensors can't resolve.

Photogrammetric techniques use overlapping imagery to reconstruct 3D surface geometry. Digital photogrammetry processes satellite or aerial stereo pairs, while Structure-from-Motion (SfM) achieves similar results using multiple overlapping photos from consumer-grade cameras, making 3D modeling far more accessible.

Interpreting remote sensing data

Digital image processing and enhancement

Raw remote sensing data needs correction and enhancement before analysis. The main processing steps are:

1. Radiometric correction adjusts for atmospheric effects and sensor calibration. Dark object subtraction removes atmospheric scattering, and converting raw digital numbers to reflectance values enables quantitative comparison across dates and sensors.
2. Geometric correction fixes spatial distortions. Orthorectification uses a DEM to remove terrain-induced distortions, and image registration aligns multi-temporal datasets so you can compare them pixel by pixel.
3. Image enhancement improves feature visibility. Contrast stretching expands the range of displayed pixel values, and band ratioing highlights spectral differences between materials (e.g., vegetation indices like NDVI, or mineral ratios for lithological mapping).
4. Spectral analysis extracts deeper information from multiple bands. Principal Component Analysis (PCA) reduces data dimensionality while enhancing subtle features. Spectral unmixing estimates the proportions of different materials within a single mixed pixel.

Classification and feature extraction

Once data is corrected and enhanced, you classify it to turn continuous spectral data into meaningful categories.

Supervised classification requires you to provide training samples (pixels you've already identified). The algorithm then assigns all remaining pixels based on those examples:

• Maximum Likelihood classifiers assign pixels based on statistical probability distributions.
• Support Vector Machine (SVM) classifiers handle complex, non-linear class boundaries more effectively.

Unsupervised classification groups pixels into spectral clusters without prior training data:

• K-means iteratively refines cluster centers until stable groups emerge.
• ISODATA adjusts the number of clusters during processing, splitting and merging as needed.

Object-Based Image Analysis (OBIA) takes a different approach. Instead of classifying individual pixels, it first segments the image into meaningful objects (groups of similar pixels), then classifies those objects using spectral, spatial, and contextual rules. This often produces cleaner results for landform mapping.

Machine learning methods are increasingly common. Random Forest classifiers handle high-dimensional data well and resist overfitting. Convolutional Neural Networks (CNNs) excel at recognizing spatial patterns, making them effective for automated landform identification in imagery.

Geomorphological analysis and integration

Change detection monitors how landscapes evolve over time. Image differencing quantifies pixel-level changes between two dates, while post-classification comparison identifies specific land cover transitions (e.g., vegetated slope to bare landslide scar).

DEM analysis extracts terrain parameters that are fundamental to geomorphology:

• Slope and aspect calculations reveal topographic characteristics
• Flow accumulation algorithms delineate drainage networks and watershed boundaries

GIS integration enhances interpretation by combining remote sensing data with other spatial datasets. Overlay analysis can merge satellite imagery with geological maps or soil data, and spatial statistics quantify patterns in geomorphological features.

Multi-sensor data fusion combines the strengths of different systems. For example, pairing a LiDAR-derived DEM (precise elevation) with multispectral imagery (surface composition) produces richer landform classifications than either dataset alone. Similarly, integrating SAR data with optical imagery provides both all-weather capability and spectral detail.

Advantages vs limitations of remote sensing

Advantages in geomorphological studies

• Large spatial coverage enables regional to global-scale analysis. A single satellite scene can cover an entire mountain range, and consistent methodology allows comparison across different regions.
• Repeated observations facilitate monitoring over time. Time series analysis reveals long-term trends like glacial retreat or desertification, and rapid response imaging captures the aftermath of extreme events like floods or earthquakes.
• Access to remote or dangerous areas expands what's possible. Active volcanoes, polar ice sheets, and vast desert regions can all be studied without physical presence.
• Non-invasive data collection minimizes disturbance to sensitive environments like coral reefs, permafrost regions, or protected habitats.
• Multi-scale analysis links fine-scale processes (soil erosion on a single hillslope) to broad landscape patterns (tectonic uplift across a region) within a single analytical framework.

Limitations and challenges

• Spatial resolution constraints mean small features like minor sinkholes or subtle fault scarps may go undetected. Mixed pixels in coarser imagery can also lead to misclassification.
• Temporal resolution gaps can miss rapid events. If a satellite revisits every 16 days, a flash flood that occurs between passes won't be captured. Cloud cover can also block optical sensors during critical moments.
• Atmospheric and environmental interference affects data quality. Atmospheric correction methods introduce their own uncertainties, and dense vegetation canopy can obscure the underlying terrain.
• Limited subsurface capability is a significant constraint. Most remote sensing techniques observe only the surface or very near-surface. Deeper structural investigation requires integration with geophysical methods like seismic surveys.
• Processing complexity demands specialized expertise. Big datasets require advanced software and computing resources, and accurate interpretation draws on knowledge spanning physics, geology, and computer science.

Practical considerations

• Cost varies widely. High-resolution commercial satellite imagery can be expensive for large areas, and airborne LiDAR surveys require significant investment. However, free data from Landsat and Sentinel has made many applications accessible.
• Validation and ground-truthing are essential. Field surveys provide in-situ measurements for calibrating and verifying remote sensing results. Combining traditional fieldwork with remote sensing consistently produces the most reliable outcomes.
• Ethical and legal issues include privacy concerns with high-resolution imagery of populated areas and international regulations governing satellite data distribution.
• Technological advancement continues to expand what's possible. Sensor technology keeps improving spatial and spectral resolution, while cloud computing and AI are transforming how large datasets are processed and analyzed.