What is the difference between hazard, susceptibility, and risk?

Susceptibility is the spatial propensity for a hazard given terrain conditions; hazard adds magnitude and probability over time; risk combines hazard with exposure (people, assets) and vulnerability (how badly they are affected). They are distinct layers and should never be relabelled as one another.

How do you build a terrain risk map in GIS?

Assemble conditioning factors (slope, aspect, geology, land cover, distance to drainage), normalise them to a common CRS and grid, weight them, and combine into a susceptibility surface. Then overlay exposure and vulnerability to move from susceptibility toward risk.

What is a weighted overlay and when should you use it?

A weighted overlay reclassifies several factor layers onto a common scale, multiplies each by a weight reflecting its influence, and sums them. It is the standard transparent method for combining terrain conditioning factors into a susceptibility index when you lack data for a fully statistical model.

What data do you need for terrain risk mapping?

A DEM and its derivatives (slope, aspect, curvature, drainage), geology and soils, land cover, an inventory of past events for validation, and exposure layers (buildings, roads, population) if you want risk rather than susceptibility.

Terrain Risk Mapping Basics

Three words that are not synonyms

Most flawed risk maps go wrong at the vocabulary stage. Susceptibility is the spatial likelihood that a hazard could occur somewhere, given the static terrain conditions — it answers where. Hazard adds magnitude and temporal probability — where, how big, how often. Risk multiplies hazard by exposure (the people and assets present) and vulnerability (how badly they would be affected) — what is actually at stake. A susceptibility map labelled "risk map" has skipped two whole dimensions. Terrain risk mapping basics start here: know which of the three you are producing, and say so. This article walks through building a susceptibility surface from terrain and then extending toward genuine risk.

The conditioning factors

Terrain susceptibility (for landslides, erosion, or flooding-related processes) is modelled from conditioning factors — the stable terrain attributes that predispose ground to a hazard. The usual set, mostly derived from a DEM, is:

Slope — the dominant control for mass movement. Derive in degrees: gdaldem slope dem_utm.tif slope.tif -alg Horn.
Aspect — proxies for moisture, vegetation, and freeze-thaw exposure.
Curvature — concave hollows concentrate water; convex spurs shed it.
Distance to drainage / streams — toe erosion and saturation.
Geology and soils — material strength and permeability.
Land cover — vegetation root reinforcement versus bare or cleared ground.
Topographic wetness index — where water accumulates.

All raster factors must be reprojected to a single metric CRS (e.g. EPSG:32633) and resampled to a common cell size before they can be combined.

Weighted overlay: the workhorse method

When you do not have a large inventory of past events to train a statistical model, the transparent default is a weighted linear overlay:

Reclassify each factor onto a common ordinal scale (say 1–5, low to high contribution). For slope you might map 0–5 degrees → 1, up to >35 degrees → 5; for distance to drainage, near → high.
Assign weights that sum to 1, reflecting each factor's relative influence (slope usually highest). The Analytic Hierarchy Process is a structured way to derive weights from pairwise expert comparison, and it produces a consistency ratio so you can check the weights are internally coherent.
Combine: susceptibility = Σ (weight × reclassified factor). Use the QGIS Raster calculator, the SAGA weighted overlay, or ArcGIS Weighted Overlay.

# QGIS raster calculator, illustrative weights:
0.45*"slope_rc" + 0.20*"geology_rc" + 0.15*"wetness_rc" +
0.10*"landcover_rc" + 0.10*"dist_drain_rc"

Where you do have an event inventory, statistical methods (frequency ratio, weights-of-evidence, logistic regression) derive the weights from data rather than expert judgement and let you quantify performance.

From susceptibility to risk

To go beyond where a hazard is likely:

Hazard: combine susceptibility with magnitude/frequency information — for example, return-period flood depths from a hydraulic model, or landslide recurrence from a dated inventory.
Exposure: overlay assets — buildings, roads, lifelines, population grids.
Vulnerability: assign damage functions or vulnerability classes to those assets.

Risk is then, conceptually, hazard × exposure × vulnerability, evaluated cell by cell or per asset. Keep the components as separate layers so each can be inspected and revised; a single fused "risk" raster with no traceable inputs cannot be defended.

Validation against an inventory

A susceptibility map is only credible if it predicts known events. With an inventory of past landslides (or flood extents), the standard check is the ROC / AUC approach: rank cells by predicted susceptibility, and measure what fraction of known events fall in the highest-susceptibility classes. An AUC near 0.5 is no better than chance; values approaching 0.8 and above indicate a useful model. Always hold back some events for validation rather than fitting and testing on the same data.

Worked example: a regional landslide susceptibility screen

Reproject GLO-30 to EPSG:32633; derive slope, curvature, and topographic wetness index.
Add geology and land cover; compute distance to drainage with the QGIS Proximity tool.
Reclassify each factor to 1–5.
Derive weights by AHP (slope dominant), check the consistency ratio.
Combine in the raster calculator into a 1–5 susceptibility surface.
Overlay the landslide inventory; compute the success-rate curve and AUC.
Symbolise in five classes, masking out flat valley floors, and document every weight and breakpoint.

Common pitfalls and why they happen

Calling susceptibility "risk" — it omits exposure and vulnerability and overstates what the map says.
Slope in the wrong units or on a geographic CRS — distorts the single most important factor.
No validation — without an inventory check, weights are unaccountable opinion.
Mismatched grids — combining factors at different cell sizes or CRSs silently misaligns them.
Over-smoothing or over-resolving — a 90 m DEM blurs the slopes that drive shallow failures; a noisy 1 m DEM injects spurious steepness. Match resolution to the process.

QA and validation checklist

All factor layers share CRS, extent, and cell size.
Reclassification breakpoints and weights are documented and justified.
Susceptibility validated against a held-out event inventory with a reported AUC.
The legend states clearly that the product is susceptibility, hazard, or risk — not loosely "risk".
Source, date, and scale recorded for every input.

Bathyl perspective

We keep hazard, susceptibility, and risk as separate, validated layers, and we publish the weights and the inventory check alongside the map. A terrain risk screen earns trust by being honest about which question it answers and by showing it predicts the events that already happened.