What hazard layers belong in an early-project GIS screening?

Typically: flood zones (e.g. FEMA NFHL where applicable), landslide susceptibility, slope from a DEM, seismic hazard (USGS or national models), liquefaction and karst where relevant, plus exposure layers like buildings and infrastructure. Start from authoritative published sources before any modelling.

What is the difference between hazard and risk in screening?

A hazard is a potentially damaging process (a flood, a landslide). Risk combines the hazard with exposure (what is there) and vulnerability (how badly it would be affected) and the consequences. A screening usually maps hazard and exposure; it does not produce a quantified risk figure.

How do I run a multi-hazard overlay in GIS?

Normalise every layer to a common CRS and resolution, reclassify each hazard to a comparable scale, then combine with a weighted overlay or simple flag. In QGIS use Reclassify by Table then Raster Calculator; keep observed events, susceptibility models and regulatory zones in separate, clearly labelled layers.

When does GIS screening need a specialist hazard assessment?

Whenever the output informs engineering design, permitting, insurance, life safety, or public communication. Screening flags where to look; site-specific geotechnical, hydrological or seismic studies are required before any of those decisions.

Natural Hazard Screening for Early Projects

Short answer

Natural-hazard screening for an early-stage project is a GIS overlay that flags, quickly and cheaply, which natural processes could constrain a site or corridor before anyone commits to it. It pulls together authoritative hazard layers — flood, landslide, slope, seismic, and the local specialities like karst or liquefaction — normalises them to a common grid, and shows where they coincide with the proposed footprint. Done well it saves expensive surprises; done badly it manufactures false confidence. The defining discipline is knowing exactly where screening stops and a specialist study must begin.

Hazard is not risk

The two words get used interchangeably and the conflation causes real harm. A hazard is a process with the potential to cause damage: a 1-percent-annual-chance flood, a debris flow, ground shaking at a given intensity. Risk is what happens when that hazard meets something exposed and vulnerable — it is a function of hazard, exposure, and vulnerability, and it carries consequences (lives, cost, downtime).

A screening map almost always characterises hazard and overlays exposure (the project footprint, nearby buildings, lifelines). It rarely characterises vulnerability or produces a defensible quantified risk number, because that needs engineering judgement about the specific assets. Saying so on the map — "hazard and exposure screening, not a risk assessment" — is not a disclaimer for lawyers; it is an accurate description of what the analysis can support.

The layers worth assembling

Start from authoritative, published sources before reaching for your own models. The exact catalogue depends on geography, but the structure is consistent:

Flooding. Where it exists, the regulatory product: in the US, FEMA's National Flood Hazard Layer with its Special Flood Hazard Areas (the 1-percent-annual-chance, "100-year" zone) and the 0.2-percent zone. Elsewhere, the national flood mapping equivalent. Supplement with a height-above-nearest-drainage (HAND) layer from a DEM where formal maps are coarse or absent.
Landslide and slope instability. Published landslide-susceptibility and inventory maps (e.g. USGS landslide products), plus a slope raster you derive yourself from a DEM.
Seismic. A seismic hazard model giving peak ground acceleration at a return period (USGS National Seismic Hazard Model in the US, or national/regional equivalents), and liquefaction susceptibility where soils and water table warrant it.
Local geohazards. Karst/sinkhole, expansive soils, coastal erosion and sea-level rise (e.g. NOAA viewers), volcanic, permafrost — include whichever the setting demands.
Exposure. The project footprint, plus buildings, roads, utilities, and protected areas to express what coincides with each hazard.

Treat observed events (an inventory of past landslides), susceptibility models (a statistical surface), and regulatory zones (a legal flood boundary) as three distinct layer categories. Merging them into one "hazard" symbol destroys the user's ability to weigh certainty.

Deriving slope correctly

Slope is the one hazard layer you usually compute yourself, and it is easy to get subtly wrong.

The single biggest error is computing slope on a DEM in geographic coordinates (EPSG:4326). There, horizontal units are degrees and vertical units are metres, so the slope algorithm divides metres by degrees and returns nonsense. Reproject to a projected CRS in metres first:

gdalwarp -t_srs EPSG:32633 -r bilinear srtm_geographic.tif dem_utm.tif
gdaldem slope dem_utm.tif slope_deg.tif -compute_edges

gdaldem slope defaults to degrees; add -p for percent slope. In QGIS the equivalent is the "Slope" algorithm under Raster terrain analysis, and in ArcGIS Pro the Spatial Analyst Slope tool — both assume planar, metric coordinates. Also record the DEM's resolution: a slope from a 30 m SRTM grid smooths away the short, steep failures that a 1 m lidar DEM would reveal, so the source resolution is part of the result's meaning.

A worked multi-hazard overlay

Goal: flag candidate constraints along a proposed corridor.

Standardise the grid. Reproject every layer to one projected CRS and resample to a common cell size (say 10 m) with gdalwarp -tr 10 10 -t_srs EPSG:32633. Mixing resolutions and CRSs is the most common source of misaligned overlays.
Reclassify each hazard to a comparable scale, e.g. 0-3. Slope: 0 for <5 degrees, up to 3 for >25 degrees. Flood: 3 inside the 1-percent zone, 1 in the 0.2-percent zone, 0 elsewhere. Use QGIS "Reclassify by table" or gdal_calc.py.
Combine deliberately. A weighted sum (0.4*flood + 0.4*landslide + 0.2*slope) gives a graded surface; a simple maximum or a boolean flag ("any hazard class >= 2") gives a conservative screen. State which you used and why — the choice is an assumption, not a fact.
Intersect with the footprint. ogr2ogr or QGIS "Zonal statistics" to summarise the hazard surface within the corridor buffer.
Keep the inputs visible. The final map should let a reviewer toggle each source layer, not just see the combined score.

The weights are subjective and must be labelled as such; presenting a weighted-overlay score as if it were measured is the screening equivalent of false precision.

QA and validation

Confirm CRS and resolution alignment before any overlay — check that two layers snap to the same grid origin.
Validate slope CRS. Spot-check a few cells; if a 45-degree hillside reads as 0.0004, the DEM was in degrees.
Check map currency and scale. A flood map can be decades old; a regional susceptibility model may be 1:250,000 and meaningless at a building footprint.
Test the edges of zones, where a footprint clips a boundary, not just clearly-in or clearly-out cases.
Reconcile categories. Make sure observed events, models, and regulatory zones are still distinguishable in the final product.

Common pitfalls and why they happen

Calling a screening a risk assessment. It maps hazard and exposure, not vulnerability or consequence; the label oversells what was done.
Slope on a geographic DEM. Mixed horizontal/vertical units make the algorithm divide metres by degrees; reproject to metres first.
Mixing official zones with informal models unlabelled. A regulatory flood line and a statistical susceptibility surface carry very different authority; merging them erases that.
Ignoring data age and source scale. A 1:250,000 layer over-zoomed to a site implies precision the source never had.
Publishing high-consequence maps without review boundaries. Anything touching safety, permitting, or insurance needs a stated escalation path to a specialist.

Bathyl perspective

We build hazard screenings as fast, honest filters: they tell a project team where to look harder, and they say plainly what they cannot decide. Every layer carries its source, date, and scale, observed events stay separate from models, and the deliverable names the threshold at which a geotechnical, hydrological, or seismic specialist must take over. Early clarity, transparent limits, a clean handoff.