What data layers are needed for flood risk mapping in GIS?

A bare-earth DEM for terrain, a hydrography network (NHD or equivalent), flow accumulation and a derived height-above-nearest-drainage (HAND) raster, regulatory zones such as FEMA SFHA, land cover for roughness, and exposure layers (buildings, population, critical assets). Each must share a projected CRS and documented vertical datum before overlay.

What is the difference between flood hazard and flood risk on a map?

Hazard describes the physical process and its probability, for example the 1% annual chance (100-year) inundation depth. Risk is hazard combined with exposure and vulnerability: who and what is in the water, and how badly they are harmed. A depth grid is hazard; expected annual damage is risk.

What DEM resolution do I need for flood screening?

Lidar-derived DEMs at 1 m or 3 m (such as USGS 3DEP) are strongly preferred for floodplain work because flood extent is governed by sub-meter vertical differences. A 30 m DEM like SRTM is only adequate for regional screening, not for parcel-level decisions.

Can GIS flood mapping replace an official FEMA flood assessment?

No. GIS screening identifies candidate exposure quickly, but insurance, permitting, and engineering decisions in the US require the effective FEMA FIRM and, where it matters, a hydraulic study and a Letter of Map Amendment or Revision.

Flood Risk Mapping: GIS Data Layers Guide

The short answer

A useful GIS flood risk map is an overlay of four distinct layer families: terrain (a bare-earth DEM and its hydrologic derivatives), hydrography (the stream and waterbody network), regulatory hazard (effective FEMA flood zones or the national equivalent), and exposure (buildings, infrastructure, population, and critical facilities). The terrain and hydrography layers let you model where water goes; the regulatory layer tells you what the authorities have already mapped; the exposure layer converts a hazard footprint into something you can actually rank and prioritise. The single most common mistake is stopping at the hazard footprint and calling it "risk."

Hazard is not risk

These words are used loosely in conversation and precisely in flood engineering. Hazard is the physical process and its probability of occurrence: the 1% annual exceedance probability (AEP) flood, still widely called the 100-year flood, with an associated inundation extent and depth grid. Risk is hazard combined with exposure (the assets located in the hazard footprint) and vulnerability (how much damage a given depth causes those assets, expressed through depth-damage functions).

The practical consequence: a 1.2 m depth grid over an empty field and the same depth grid over a hospital are identical hazards but vastly different risks. If your deliverable colours the floodplain and stops, you have made a hazard map. To make a risk map you must intersect that footprint with exposure and weight it by consequence. Label every output with which of the two it actually shows, because downstream readers will assume the stronger claim if you let them.

The terrain layer: a bare-earth DEM and its derivatives

Floodplain extent is governed by elevation differences that are often smaller than a metre, so vertical accuracy dominates everything. Use a bare-earth (DTM) product, not a digital surface model that includes tree canopy and rooftops, which would dam the flow artificially.

Resolution. Lidar-derived DEMs such as USGS 3DEP at 1 m or 1/3 arc-second (about 10 m) are the working standard. The global 30 m SRTM/Copernicus DEMs are acceptable only for continental screening. At 30 m a single cell can hide a levee, a road embankment, or a 1.5 m drop that controls the whole inundation pattern.
Vertical datum. Confirm it explicitly. US lidar is typically NAVD88 in metres or US survey feet; older data may be NGVD29. Mixing a NAVD88 DEM with NGVD29 flood elevations introduces a regionally varying error of tens of centimetres, which is enough to flip a parcel in or out of the floodplain.
Hydro-conditioning. Raw DEMs contain spurious pits. Before deriving flow, fill or breach them. In a tool like Whitebox, BreachDepressionsLeastCost is usually preferable to a naive fill because it carves drainage through embankments instead of flooding flat areas. In GDAL terms you would still warp everything to a common projected grid first with gdalwarp -t_srs EPSG:6350 -tr 1 1 -r bilinear (NAD83(2011) UTM is a good analytical choice in the US).

From the conditioned DEM you derive flow direction, flow accumulation, and the channel network. The most useful single screening derivative is HAND (Height Above Nearest Drainage): for every cell, the vertical distance to the nearest cell on the stream network along its flow path. A HAND raster thresholded at, say, 2 m, 5 m, and 10 m gives a fast, physically meaningful first-pass inundation susceptibility without running full hydraulics.

The hydrography layer

You need the stream and waterbody network the terrain will be referenced to. In the US the National Hydrography Dataset (NHD/NHDPlus HR) provides flowlines, catchments, and pre-computed flow accumulation; the European equivalent is the EU-Hydro and Copernicus river network. Snap your DEM-derived channels to this authoritative network rather than trusting the DEM alone, because lidar over wide rivers and culverts is unreliable. NHDPlus also carries the catchment attributes used by the NOAA National Water Model, which is the operational source for streamflow forecasts you may want to join.

The regulatory hazard layer

In the US this is the FEMA National Flood Hazard Layer (NFHL), served as both a web service and downloadable geodatabase. The fields that matter:

Special Flood Hazard Area (SFHA) — zones A and AE, the 1% AEP floodplain that triggers mandatory insurance for federally backed mortgages.
Zone X (shaded) — the 0.2% AEP (500-year) floodplain.
Base Flood Elevation (BFE) — the modelled water-surface elevation of the 1% AEP flood, the number against which finished-floor elevations are compared.

Treat the NFHL as authoritative for regulatory status but understand its limits: many panels are decades old, approximate Zone A areas have no BFE, and the maps do not model pluvial (rainfall-driven, off-channel) flooding at all. That gap is exactly where your HAND and pluvial screening add value.

The exposure layer

This is what turns hazard into risk. Sources include OpenStreetMap or Microsoft/Google building footprints, the national address point file, gridded population (WorldPop, GHS-POP), parcel data, and critical-facility inventories (HIFLD in the US: hospitals, fire stations, power, water treatment). For a defensible risk score you also need depth-damage functions — the FEMA Hazus library and the USACE depth-damage curves relate inundation depth at a structure to a percentage of replacement value lost.

A worked screening workflow

Acquire a 1 m 3DEP DEM tile and the NHDPlus HR flowlines for the basin.
Reproject both to a single projected CRS and grid: gdalwarp -t_srs EPSG:6350 -tr 1 1 -r bilinear dem.tif dem_utm.tif.
Hydro-condition the DEM (breach depressions), then compute D8 flow direction and accumulation.
Generate the HAND raster from the conditioned DEM and the snapped stream network.
Reclassify HAND into susceptibility bands (≤2 m high, 2–5 m moderate, 5–10 m low).
Load the FEMA NFHL and clip the SFHA and BFE polygons to the study area.
Intersect building footprints with both the HAND bands and the SFHA: ST_Intersects in PostGIS, or the QGIS "Extract by location" / "Join attributes by location" algorithms.
For each exposed structure, estimate depth (BFE minus finished-floor elevation, or a HAND-based proxy) and apply a depth-damage curve to produce a relative loss estimate.
Aggregate to a reporting unit (parcel, census block, asset class) and rank.

The output is now a risk screening, not a coloured floodplain.

Common pitfalls and why they happen

Using a DSM instead of a DTM. Canopy and buildings act as dams in the flow model, so flood pathways stop where they should continue. The failure is silent because the map still renders.
Ignoring the vertical datum. Combining elevations from different vertical references produces errors of the same magnitude as the depths you are trying to map.
Mistaking the SFHA for total flood risk. FEMA maps the riverine 1% AEP; they do not capture pluvial flooding, dam failure, or events larger than the base flood. Properties just outside Zone X still flood.
Buffering or measuring in degrees. If the DEM or footprints are still in EPSG:4326, every distance, area, and cell-size assumption is wrong. Reproject first.
Treating one map vintage as current. NFHL panels and lidar tiles have effective dates years apart; document both.

Validation

Cross-check the modelled extent against an observed flood event where one exists — Sentinel-1 SAR-derived inundation (open water is dark in VV polarisation) and the Copernicus Emergency Management Service rapid-mapping products are good independent references. Confirm the HAND channel network aligns with NHD flowlines. Spot-check a handful of finished-floor elevations against field survey if any exist. And state the limitations in plain language: resolution, datum, the AEP modelled, and what is excluded (pluvial, coastal surge, future climate).

Bathyl perspective

We treat flood layers as a stack with explicit provenance: every raster carries its source, date, CRS, vertical datum, and AEP in the metadata, and hazard outputs are never relabelled as risk. The deliverable is a screening that tells a planner where to look first and a clear note on where it must hand off to a hydraulic engineer and the effective FIRM.

Flood Risk Mapping Data Layers