Data and map guide

Data sources

Every city is modelled from the same kinds of input, drawn from official and open sources for its country. The exact providers, collection dates and licences differ from city to city.

• Terrain. A bare-earth digital terrain model at 2 m spacing from national or municipal lidar programmes, such as USGS 3DEP in the United States, the NRCan High Resolution DEM or a city’s own lidar survey in Canada, Geobasis NRW DGM1 in Germany and Digitaal Vlaanderen DHM Vlaanderen II in Belgium.
• Land cover. Classified from Sentinel-2 satellite imagery at 10 m into Dynamic World classes (built area, trees, grass, crops, bare ground, water and others), composited over several cloud-free scenes for each city. Where a city or country publishes higher-resolution land cover, that dataset is used instead.
• Roads and watercourses. Official road and hydrography networks, such as Census TIGER/Line and the USGS National Hydrography Dataset, ATKIS, the Flemish Wegenregister and VMM hydrography, and national or municipal mapping in Canada. Permanent water bodies come from OpenStreetMap.
• Soils. Global hydrologic soil groups at 250 m (HYSOGs250m), which govern how much rainfall soaks in rather than running off.
• Design rainfall. National rainfall statistics for each return period and duration: Environment and Climate Change Canada IDF curves, NOAA Atlas 14 in the United States, DWD KOSTRA in Germany and KMI/IRM in Belgium.
• Boundaries. Official municipal boundaries, such as Statistics Canada census subdivisions, Census TIGER/Line places and Eurostat GISCO municipalities, with a short buffer around each for context.

Methodology

The maps are produced by an in-house, GPU-accelerated simulation workflow written in Rust. It implements a new approach developed by GeoRetina that is significantly faster than the conventional method we used before while matching its results almost everywhere, which is what makes twenty scenarios for a hundred cities practical. Every scenario for every city goes through the same steps:

1. Terrain preparation. The lidar terrain is cleaned, clipped to the city with a buffer for context, and conditioned so that water can move along real surface pathways such as streets, channels and culverted crossings.
2. Rainfall to runoff. For each scenario the design storm is spread over its duration, and the share that infiltrates is estimated from land cover and soil groups; the rest becomes surface runoff.
3. Surface flow. Runoff is routed across the terrain as two-dimensional shallow-water flow on the GPU, cell by cell at the model resolution, for the length of the storm and the drainage that follows it.
4. Results. The maximum water depth reached in every cell is kept, resampled to the city’s published grid, and classified into the depth ranges and risk classes shown in the legend. Depths below the display threshold are hidden.

The simulation represents surface water only. It does not include storm sewers and their inlets, river or coastal flooding, groundwater, or the interiors of buildings, which is why the maps are screening products rather than official designations.

Scenarios are design rainfall, not forecasts

A scenario combines a rainfall return period of 10, 25, 50 and 100 years with a storm duration of 1, 3, 6, 12 and 24 hours. A complete city release contains all 20 combinations. Each map shows potential surface-water flooding under the selected rainfall scenario.

A return period describes the statistical rarity of the rainfall event that was simulated. It is not the probability that a particular property floods, and it says nothing about river, coastal or groundwater flooding, or about a specific past or future storm.

Two layers, one scenario at a time

Depth ranges show the simulated maximum water depth reached during the scenario, grouped into classes. Risk classification shows the potential flood-hazard category. The two layers are never shown together, so one opaque layer cannot hide the other. The standard legend is shown below; each city page provides the legend for its selected release.

• Depths below 0.05 m are hidden in both depth and risk views. Depths at or above this threshold are displayed where data are available, except within the permanent-water mask.
• Transparent areas can indicate shallow depths below the display threshold, mapped open water excluded from display, or unavailable data. Transparency does not mean an area is dry or safe.
• Risk class 0 (“Very Low Risk”) is the lowest modelled class, not a guarantee that flooding cannot occur.

Permanent water masking

We hide mapped permanent open water, including lakes, reservoirs, river channels, canals and harbours, from both the depth and risk overlays using OpenStreetMap water polygons. This helps distinguish existing water bodies from flooding on normally dry land. The mask changes only the displayed maps; the underlying simulation results are preserved.

Only the mapped water footprint is hidden, with no added buffer. Simulated inundation outside that footprint remains visible. Areas explicitly tagged as seasonal or intermittent water, or as wetlands, are excluded from the mask. OpenStreetMap boundaries and tags can be incomplete or outdated, so the mask is an approximation of permanent water coverage.

Masked water should be interpreted as existing water where flood hazard is not assessed by this display, not as zero depth or no risk. These rainfall scenarios do not provide a separate assessment of river or coastal flooding.

Map detail

The maps show results on a grid of 5–30 m, depending on the city’s source data; each city page states its own spacing. Zooming in further enlarges the same cells; it does not add detail or precision. Resolution is not a guarantee of accuracy. Local conditions, obstructions and changes since the data were collected can affect actual flooding.

Exact numeric depths are not part of the public maps. Clicking the map shows the selected coordinates together with the legend so a location can be interpreted against the class it falls in.

Coverage

Each city is modelled over a defined area, noted on its map page. Outside that area the map is transparent: no result, rather than zero depth or low risk.

Releases and citations

Each city’s maps are published as an immutable release with an identifier such as 2026-09-public-v1, shown on the map page. A release never changes after publication; corrections and updates are always published as a new release. Cite the release identifier with any map you reference.

A shared link opens the release it was created from, so two people looking at the same link see the same maps. When a newer release exists, the page says so and offers it. If a release has been withdrawn, the link explains that and offers the latest release instead of quietly showing different results.

Updates and new cities

There is no fixed schedule. A city’s maps are updated when we find an issue, when better source data becomes available, or when feedback shows something is wrong or unclear. Every update is a new release, so the previous maps stay citable and a shared link keeps showing what it showed.

New cities are added on request. A request form is coming; until then, ask through the GeoRetina AI platform. Requests are prioritised by interest and by whether the source data a city needs, mainly terrain and rainfall statistics, is available at the quality the maps require.

Basemap, terrain and attribution

The background map is built from OpenStreetMap data (© OpenStreetMap contributors, ODbL) using the open Protomaps basemap and is served by GeoRetina; there is no aerial imagery. The optional 3D view draws terrain from Esri elevation services and buildings and streets from OpenStreetMap. The flood overlay itself is produced by GeoRetina; each city page lists the data sources and licences that apply to that release.

Appropriate use

These maps are screening-level information for exploring how modelled surface water behaves under design storms. They are not official flood hazard designations, insurance ratings or engineering assessments and must not replace them. For decision-grade analysis, custom scenarios and reports, see the GeoRetina AI platform.