What data layers do you need for landslide susceptibility screening?

At minimum: a DEM-derived slope and aspect, surficial and bedrock geology, a hydrology layer (distance to streams, drainage density or a wetness index), land cover, and a landslide inventory for calibration and validation. Triggering factors such as rainfall intensity or seismic shaking refine the result.

Is slope alone enough to map slope stability?

No. Slope angle is the dominant control but failures depend on material strength, water pressure, structural orientation and triggers. A slope-only map will misclassify strong rock cliffs as hazardous and weak clay slopes as safe.

What slope angle is considered high risk for landslides?

It is material-dependent, but shallow translational failures cluster broadly between about 25 and 45 degrees. Below roughly 10 degrees driving stress is usually low; above 45 degrees in soil, debris tends to have already shed. Always calibrate thresholds against a local inventory.

How do you validate a landslide susceptibility map?

Withhold part of the landslide inventory, then compute a success-rate or ROC curve and the area under the curve (AUC). An AUC above about 0.75 is usually considered acceptable; below 0.7 the model is weak.

Slope Stability GIS Screening: A Practical Workflow

Short answer

A slope stability screening layer ranks terrain by relative likelihood of slope failure so planners can flag candidate problem areas before commissioning geotechnical investigation. It is built by combining a slope raster with geology, hydrology, land cover and a landslide inventory, then weighting those factors against where failures have actually occurred. It is decision support, not a factor-of-safety calculation, and it never replaces a site-specific geotechnical assessment.

The single most common mistake is treating it as a slope-only product. Slope angle is the strongest predictor, but material strength, groundwater, structural geology and triggers (rainfall, seismicity, undercutting) determine whether a steep cell actually fails.

What "susceptibility" means and what it does not

Three terms are routinely confused. Susceptibility is the spatial likelihood of failure given the terrain's intrinsic properties, with no time component. Hazard adds a temporal probability and magnitude (e.g. probability of a debris flow of a given volume in 50 years). Risk multiplies hazard by exposure and vulnerability of people and assets. A GIS screening exercise almost always produces susceptibility, occasionally hazard if a triggering scenario is included, and rarely true risk. Label your output accordingly, because a "susceptibility" raster presented as a "risk map" will be misread by every non-specialist who sees it.

The reason this matters operationally: susceptibility tells you where to look, not when failure happens or what the consequence is. A high-susceptibility cell on an uninhabited ridge is low risk; a moderate-susceptibility cell above a school is not.

The input layers, with the parameters that matter

Slope and curvature from the DEM

Compute slope in degrees from the best available DEM. For regional screening a 10–30 m DEM (SRTM 1-arc-second, Copernicus GLO-30) is adequate; for detailed work use a 1–2 m lidar DEM. Slope is sensitive to cell size: a 30 m DEM smooths short steep faces and will systematically under-report slope on small landslide scars, while a 1 m lidar DEM amplifies micro-relief and noise. Pick a resolution that matches the failure size you care about.

In QGIS use Slope (from gdaldem slope) in the Raster terrain analysis group; in GDAL directly:

gdaldem slope dem.tif slope_deg.tif -compute_edges -p

Use -p for percent or omit it for degrees, and -s 111120 only if the DEM is in geographic (degrees) coordinates so horizontal and vertical units differ. Mismatched units here are the classic silent error: running slope on a metre-elevation DEM stored in EPSG:4326 without a scale factor produces nonsense.

Profile and plan curvature add value: failures concentrate on concave hollows where water and colluvium accumulate. Derive curvature with gdaldem is not available, so use QGIS Curvature or SAGA Slope, aspect, curvature.

Geology

Surficial geology controls material strength. Reclassify lithology into susceptibility classes: weak, weathered or clay-rich units (mudstone, schist, thick colluvium, glacial till) rank high; massive intact crystalline rock ranks low. Where structural data exist, dip direction relative to slope aspect matters enormously: a "dip slope" where bedding dips parallel to and out of the hillside is far more failure-prone than the same slope cut against bedding.

Hydrology

Pore water pressure is the usual trigger of shallow failure. Proxies: distance to drainage, drainage density, and the topographic wetness index (TWI = ln(a/tan β), where a is upslope contributing area per unit contour length and β is slope). Compute TWI with SAGA Topographic Wetness Index after a flow-accumulation step. Add distance to streams with gdal_proximity or QGIS Proximity (raster distance).

Land cover and an inventory

Land cover affects rooting strength and runoff; deforested slopes are notably more failure-prone. Finally, a landslide inventory — mapped scars and deposits — is non-negotiable for any defensible product. It is both the calibration target and the validation set. Use national inventories (for example USGS landslide inventories in the US) or map scars from lidar hillshade.

A worked screening workflow

Suppose you are screening a 400 km² corridor on a 10 m DEM, with a polygon inventory of 180 past failures.

Harmonise everything. Reproject all layers to one projected CRS (e.g. a UTM zone) so distances and areas are in metres, resample rasters to a common 10 m grid and snap extents. Use gdalwarp -t_srs EPSG:32633 -tr 10 10 -r bilinear for continuous layers and -r near for categorical ones.
Build factor rasters. slope_deg, curvature, TWI, distance-to-stream, geology class, land cover class — all aligned 10 m rasters.
Weight by evidence, not by intuition. The defensible route is the frequency ratio (or weights-of-evidence / information value): for each class of each factor, compute the ratio of (% of landslide pixels in that class) to (% of total area in that class). A ratio above 1 means the class is over-represented among failures. Sum the class scores per pixel to get a susceptibility index. This calibrates thresholds to your actual data instead of importing someone else's numbers.
Classify. Break the continuous index into 4–5 classes (very low to very high) using natural breaks or, better, quantiles of the index at known landslide locations.
Validate. Split the inventory 70/30 before step 3. Plot a success-rate curve (cumulative % of validation landslides captured vs cumulative % of area ranked from highest susceptibility) and report AUC. An AUC around 0.80 is good for regional screening; under 0.70 means a factor is missing or mis-weighted.

Slope thresholds, with the caveat

Where no inventory exists and a first-pass cut is needed, broadly defensible bands for shallow soil/debris failures are: under ~10° low driving stress, ~10–25° moderate, ~25–45° elevated, and above ~45° often bedrock or already-stripped. These are starting points only — rockfall sources sit above 45–60°, and slow earthflows in sensitive clays move on slopes under 10°. Never publish fixed thresholds without stating they require local calibration.

Common pitfalls and why they happen

Slope-only maps. They happen because slope is the easiest layer to make. The result over-predicts on strong steep rock and under-predicts on weak gentle slopes.
Resolution mismatch. Mixing a 30 m DEM-derived slope with a 2 m land cover layer means the slope is smoothed relative to everything else; small scars vanish. Resample to a common grid first.
Inventory leakage. Calibrating and validating on the same landslides inflates AUC. Always split before weighting.
Ignoring the trigger. A susceptibility map says nothing about when. Presenting it as a hazard map invites false confidence after a dry summer.
Edge artefacts. Without -compute_edges, slope rasters carry a NoData fringe that propagates through every overlay.

QA and validation checklist

All layers in one projected CRS, common cell size, snapped extents, NoData handled.
Slope units (degrees vs percent) verified against vertical units.
Factor weights derived from a calibration inventory, not copied.
Independent validation subset with reported AUC and success-rate curve.
Output labelled "susceptibility (relative)" with date, DEM source/resolution and the statement that geotechnical assessment is required for any design or safety decision.

Bathyl perspective

We use susceptibility screening to direct limited field and geotechnical budgets to the cells that warrant a closer look, and we ship the calibration curve and factor weights alongside the raster so the result is auditable. The map narrows the search; the engineer makes the call.

Slope Stability GIS Screening