What is a landslide inventory and what should it contain?

It is a spatial record of mapped landslides — usually polygons for source, transport and deposit zones, plus attributes for type, activity, confidence, date and detection method. It is the foundation for susceptibility and hazard mapping.

How do you map landslides in GIS using terrain data?

Combine hillshades and slope derived from a high-resolution DEM (ideally lidar), with stereo or time-series imagery, to recognise scarps, hummocky deposits and displaced ground, then digitise zones at a consistent scale.

How do you attribute and classify landslides?

Record movement type (slide, flow, fall, topple, complex), activity state, certainty, area, date or date range, and detection method, so each polygon's reliability and age are explicit.

How do you assess the completeness of a landslide inventory?

Compare mapped landslides against the frequency-area distribution, cross-check against independent imagery and field data, and document the mappable size threshold set by your DEM resolution.

GIS for Landslide Inventory Mapping

Short answer

A landslide inventory is a spatial database of mapped slope failures — where they are, what kind they are, how active they are, and how confident you are about each one. It is the single most important input to any susceptibility or hazard model, because those models are trained on or validated against the inventory. The craft lies in mapping zones (scarp, transport, deposit) rather than blobs, attributing each polygon with type, activity and confidence, and being honest about the completeness and resolution limits of the result.

What goes in the inventory

Map landslides as polygons, and where possible separate the morphological zones rather than drawing one outline:

Source / scarp: the depletion zone, marked by the main scarp.
Transport: the track or flow path.
Deposit / accumulation: hummocky, displaced material at the toe.

For small or distant features a single polygon plus a scarp line is acceptable, but record which convention each feature uses.

The attribute schema is what turns drawings into data. A workable set, broadly aligned with international guidance (Varnes / Hungr classification, Cruden & Varnes activity descriptors):

landslide_id, mapper, map_date
movement_type: fall, topple, slide (rotational/translational), spread, flow, complex
material: rock, debris, earth
activity: active, suspended, dormant, relict, stabilised
confidence: definite, probable, possible
detection_method: lidar/DEM, aerial imagery, InSAR, field, archival
area_m2, date_range_min, date_range_max
source and notes

Storing confidence and detection_method is not optional — a polygon traced from 2 m lidar is not equivalent to one inferred from a 30 m DEM, and downstream models need to weight them differently.

Mapping from terrain data

High-resolution terrain is the workhorse. Lidar-derived DEMs (0.5–2 m) reveal scarps and hummocky deposits that vegetation hides on imagery, especially in forested terrain.

Generate visualisation layers from the DEM:

gdaldem hillshade -z 1 -az 315 -alt 45 dem.tif hs_315.tif
gdaldem hillshade -z 1 -az 045 -alt 45 dem.tif hs_045.tif
gdaldem slope -p dem.tif slope_pct.tif

Make multiple hillshades from different azimuths (e.g. 315° and 045°): a single illumination direction hides features that run parallel to the light. Better still, compute a multi-directional hillshade or a slope map, which is illumination-independent and excellent for picking out steep scarps and the break in slope at deposit toes. Visualisations such as Sky-View Factor or Simple Local Relief Model (available via the SAGA/GRASS toolboxes and the RVT tool) further enhance subtle micro-topography.

Set a consistent mapping scale (for example digitise at 1:2,000–1:5,000 with a 1 m lidar base) so polygon detail is uniform. The DEM resolution sets a hard floor on the smallest mappable landslide; state it.

Detecting change and movement

Two methods add a temporal dimension:

DEM differencing: subtract a pre-event DEM from a post-event one to map volume gained and lost. With aligned, co-registered DEMs:
```
gdal_calc.py -A dem_post.tif -B dem_pre.tif --calc="A-B" \
  --outfile=dem_diff.tif --NoDataValue=-9999
```
Negative values are depletion (scarp), positive values accumulation (deposit). Co-registration error dominates the result, so align the DEMs on stable ground first.
InSAR: satellite radar interferometry (Sentinel-1, processed via ESA SNAP or services like the COMET-LiCS / EGMS products) measures millimetre-to-centimetre line-of-sight displacement over time, ideal for slow, persistent movement on dormant or creeping slides. It detects motion that morphological mapping alone would miss, and lets you populate the activity attribute with evidence rather than judgement.

Worked workflow

Assemble a lidar DEM in a projected metric CRS (national grid or UTM); build hillshades (≥2 azimuths), slope, and a relief visualisation.
Set up the polygon and scarp-line layers with the attribute schema and value lists in QGIS.
Map systematically across the area at the fixed scale, recording confidence and detection_method per feature; keep a consistent eye height/zoom.
Cross-check ambiguous features against orthoimagery and, where available, InSAR or DEM-difference.
Attribute area automatically: in QGIS field calculator $area, or in PostGIS ST_Area(geom) (metric CRS gives m²).
Validate completeness and topology before publishing.

Assessing completeness

A landslide inventory is never complete; the question is how complete, and you can quantify it. The frequency–area distribution of landslides typically follows a power law for medium and large slides but rolls over below a characteristic size. Plot the number of landslides against area on log–log axes: a clean rollover at the small end is expected and indicates the resolution limit of your data; an early or noisy rollover suggests under-mapping of medium events that you should be able to see. This is a standard, well-established diagnostic in the landslide literature and a far better completeness check than visual impression.

Also cross-validate against an independent source — a separate imagery epoch, a field survey, or another mapper's coverage of a sample area — and report the agreement.

Common pitfalls and why they happen

Single-azimuth hillshade. Features parallel to the light direction vanish, so they go unmapped. Use multiple azimuths or slope.
Blob mapping. One outline per landslide loses the scarp/deposit distinction that hazard models need. Map zones, or at least a scarp line.
No confidence or method field. Mixing lidar-certain and DEM-inferred polygons without flagging them corrupts any model trained on the inventory. Always attribute reliability.
Uncoregistered DEM differencing. Vertical offsets on stable ground masquerade as volume change. Co-register first and check the residual on flat, stable areas.
Ignoring the resolution floor. Reporting tiny landslide counts from a coarse DEM implies a precision you don't have. State the mappable threshold.
Inconsistent scale. Polygons digitised at wildly different zooms produce a heterogeneous dataset. Fix the mapping scale.

Validation and QA

Run topology checks: no overlapping landslide polygons (unless intentionally nested zones), no invalid geometry (ST_MakeValid / QGIS Fix geometries).
Confirm every polygon has movement_type, confidence and detection_method populated.
Verify areas are in m² (metric CRS) and sanity-check the largest and smallest.
Plot the frequency–area distribution and document the rollover.
Have a second mapper review a sample; record inter-mapper agreement.

Bathyl perspective

We build landslide inventories as evidence with stated confidence, not as a single authoritative outline. When each polygon carries its detection method and certainty, the inventory can be re-weighted as new lidar or InSAR arrives — and the susceptibility models built on top inherit honesty about what is known and what is inferred.