What is a geological lineament?

A lineament is a mappable linear feature on the surface that reflects underlying structure, typically faults, fractures, joints, or lithological boundaries, expressed as aligned valleys, ridges, drainage segments, or tonal contrasts.

Which imagery is best for lineament mapping?

Multi-directional hillshade from a DEM is often the strongest single source because it emphasizes topographic expression of structure. Combine it with multispectral imagery (Sentinel-2, Landsat) and radar for full coverage.

How do I extract lineaments from satellite imagery?

Enhance edges with directional (Sobel/edge) filters or multiple hillshade azimuths, then either digitize manually or run automated line-detection (e.g. LINE in PCI, or a Hough/Canny pipeline), and analyze orientations with a rose diagram.

Should I map lineaments manually or automatically?

Manual mapping uses geological judgement and avoids cultural artefacts but is subjective and slow. Automated extraction is fast and repeatable but over-detects roads, field edges, and noise. A hybrid, automated then expert-edited, is usual.

Lineament Mapping From Satellite Imagery

Short answer

Lineament mapping is the extraction of linear surface features (aligned valleys, ridges, straight drainage segments, tonal or vegetation breaks) that express underlying geological structure such as faults, fractures, and joints. The single most effective input is usually a multi-directional hillshade from a DEM, because most structures are expressed topographically, supplemented by multispectral and radar imagery. The hardest part is not detection but discrimination: a straight line on the image is just as likely to be a road, a pipeline, a field boundary, or a sun-angle artefact as a fault. Good lineament mapping therefore combines directional enhancement, multiple illumination angles, and geological judgement, ending in an orientation analysis (a rose diagram) rather than a raw line count.

What a lineament is, and what it is not

A lineament is a mappable linear feature whose alignment plausibly reflects structure at depth. Faults and fracture zones often guide erosion, so they show up as linear valleys, straight stream reaches, offset ridges, springs, or abrupt tonal and vegetation changes. Joints and bedding can produce parallel lineament sets.

Two cautions belong at the front, not the end:

A lineament is an interpretation, not an observation. You observe a linear feature; you infer it is structural. That inference needs corroboration.
Illumination bias is real and well documented. A single hillshade lit from the northwest (azimuth 315°) suppresses features that run roughly parallel to the light and over-emphasizes those perpendicular to it. Mapping from one sun angle systematically biases your orientation statistics.

Choosing inputs

Different data sources reveal different parts of the structural picture:

DEM-derived hillshade (SRTM 30 m, Copernicus 30 m, or LiDAR where available) emphasizes topographic expression. This is the workhorse for structural lineaments.
Multispectral optical (Sentinel-2 10 m, Landsat 30 m): tonal, drainage, and vegetation lineaments. False-colour band combinations and ratios can sharpen lithological and alteration boundaries that follow structure.
Radar (Sentinel-1 SAR) has its own side-looking illumination geometry, which highlights a different set of features than optical sun-lit imagery, useful as an independent check on illumination bias.

Enhancing the data so structure stands out

Multi-directional hillshade. The fix for single-azimuth bias is to compute several hillshades and combine them. Generate, for example, azimuths 315°, 045°, 135°, and 225°:

gdaldem hillshade dem.tif hs_315.tif -az 315 -alt 45
gdaldem hillshade dem.tif hs_045.tif -az 45  -alt 45

Then blend them, or use GDAL's combined multidirectional option:

gdaldem hillshade dem.tif hs_multi.tif -multidirectional

This reveals lineaments in all orientations roughly evenly, which is essential before you trust an orientation rose.

Directional and edge filters. On imagery or the DEM, convolution filters sharpen linear edges. Sobel and Laplacian kernels highlight gradients; directional (anisotropic) kernels emphasize features in a chosen orientation. In QGIS this is available through SAGA/GRASS convolution tools or the Raster Calculator with a custom kernel; in many image-processing tools a Canny edge detector is the first step before line extraction.

Band ratios (optical) such as a clay or iron-oxide ratio can make structurally controlled alteration trends visible where topography is subdued.

Manual vs automated extraction

Manual digitizing. An analyst draws lineaments on screen over the enhanced rasters. Strengths: geological context, ability to reject obvious cultural features, and continuity across subtle gaps. Weaknesses: subjective and operator-dependent (two geologists rarely produce identical maps), slow over large areas, and prone to confirmation bias toward expected trends. Mitigate by digitizing at a fixed scale and from multiple illuminations.

Automated extraction. Algorithms detect edges then fit line segments. The classic is the LINE module in PCI Geomatica; an open equivalent is a Canny-edge then Hough-transform pipeline (available via OpenCV or scikit-image), and GRASS offers extraction tools. Strengths: fast, fully repeatable, objective parameters. Weaknesses: over-detects everything linear (roads, field edges, scan lines) and has no idea what is geological. Parameters (edge threshold, minimum segment length, maximum gap to bridge, angular tolerance for merging) must be tuned and reported.

Hybrid is the practical standard: run automated extraction, then have a geologist cull cultural features and noise and refine continuity.

A worked workflow

Prepare the DEM. Reproject to a projected CRS so azimuths are geometrically meaningful: gdalwarp -t_srs EPSG:32637 dem.tif dem_utm.tif.
Generate multi-directional hillshade (commands above) plus 2–3 single-azimuth versions for cross-checking.
Mask cultural features. Bring in OpenStreetMap roads, railways, and power lines and overlay them so you can recognize and exclude man-made linears.
Extract lineaments automatically with documented parameters, or digitize manually at, say, 1:50,000, keeping line geometry as a vector layer.
Edit. Remove segments coincident with roads, field edges, and obvious artefacts.
Analyze orientation. Compute the strike of each segment and plot a length-weighted rose diagram (QGIS has plugins for this; or compute azimuths in PostGIS with degrees(ST_Azimuth(start, end)) and bin them). Length-weighting prevents many short noisy segments from dominating.
Correlate. Compare dominant lineament trends to known regional fault orientations, published structural maps, and any field measurements.

Common pitfalls and why they happen

Single-azimuth illumination bias. Mapping off one hillshade produces a rose diagram skewed perpendicular to the light direction. Always use multidirectional or several azimuths.
Counting cultural features as structure. Roads, canals, and field boundaries are dead straight and abundant; without a cultural mask they swamp the dataset and corrupt orientation statistics.
Reporting counts, not length. Lineament density and orientation should be length-weighted; raw counts overweight noise.
Treating lineaments as confirmed faults. They are candidates. Without field, geophysical, or published corroboration they remain hypotheses.
Mixing scales. Lineaments digitized at 1:25,000 and 1:250,000 are not comparable; fix the mapping scale.

Validation and QA

Cross-check the lineament set against an independent illumination source (radar, or a different hillshade azimuth) and confirm the major trends persist rather than appearing only under one light. Overlay published fault maps and regional structural trends; agreement supports a structural origin. Quantify operator subjectivity, where possible, by having the area mapped twice and comparing. Record the input data, enhancement parameters, extraction method, mapping scale, and CRS so the rose diagram is reproducible.

Bathyl perspective

We map lineaments as a hypothesis-generating layer, not a structural truth. The discipline is in the controls: multidirectional illumination to defeat sun-angle bias, a cultural-feature mask to keep roads out of the statistics, and length-weighted orientation analysis. We deliver the rose diagram alongside the assumptions behind it, so a structural geologist can judge how much of the signal is real.