What the Flood Risk Services Do

Continuuiti answers three flood questions for any site, each with its own service and its own data:

All three are available through the Flood Risk API, are documented in the API reference, and are read together. The flood field says whether rivers or the sea reach the site. The terrain shows where water can collect. The simulation shows what a storm does on that ground, with the buildings in the way. None of the three is a forecast of a particular event, and none is a site-level engineering model.

Data Sources

Attribution lines, as carried in every API response:

• Global river flood hazard maps, European Commission, Joint Research Centre (Baugh et al. 2024).
• WRI Aqueduct Flood Hazard Maps V2, World Resources Institute, CC BY 4.0.
• GEDTM30 © OpenGeoHub Foundation, CC BY 4.0.
• Google Open Buildings, CC BY 4.0, or Microsoft GlobalMLBuildingFootprints, CDLA Permissive 2.0, whichever served the site.
• © OpenStreetMap contributors, ODbL 1.0, via Overture Maps.

Licences travel with each layer. Every API response names the licence and attribution of each layer it carries, so a report built on the data can carry them forward. Images of flood grids and simulations on this site are derived from these datasets and the attributions above apply to them.

Full citations are on the Data Sources page.

Return Periods Explained

A return period describes how rare a flood is. A 100-year flood (RP100) has a 1% chance of being equalled or exceeded in any given year. It does not happen once every 100 years on a schedule: it can happen two years running and then not for two centuries.

The flood field carries these four because both published map sets include them. The simulation carries whatever storms you give it; the return period of a storm belongs to the rainfall curve you took it from.

Flood Field: Riverine and Coastal Grids

The Flood Field service reads the two published hazard maps over a square box centred on the site, 6 km across by default, at each map’s native resolution, and returns them as they are published.

Riverine. The grid holds flood depth in metres above local ground for each return period. A cell is null when it lies outside the modelled floodplain; the map holds no zero cells in practice, so null and “dry” are kept apart. The site cell’s value is reported for each return period, and nearest_real_cell_m gives the distance to the nearest cell with a modelled depth, so a site just outside the floodplain is read as such. The map also marks the river channel itself (permanent_water_class); a site that lands in the channel is identified, and its cell value is the water in the channel, which is why the band is carried.

Coastal. The grid holds coastal inundation depth in metres at the historical reference (2010 sea level, extreme sea levels fitted to 1979 to 2014). null is sea or an unmodelled cell; 0.0 is modelled dry land, a real value. Two readings are carried for every return period: without land subsidence and with it. Subsidence matters in delta and coastal cities where the ground is sinking, and the reading with subsidence is the one to use there.

Grid conventions. Row 0 is north. The origin is the centre of the north-west cell. A cell’s centre is origin.lat minus row times dy_deg, origin.lon plus col times dx_deg, and its value sits at index row times width plus col in the flat array. The angular cell steps georeference the grid; metre sizes are labels for reading.

The band audit. For every grid band the response reports how many cells hold a value, how many are masked, how many are exactly zero, how many carry a known fill value, and the minimum, maximum and median over real cells. A reader can judge a grid before mapping it: a box with ten modelled cells and a box with three thousand are different pictures. Riverine coverage should grow with return period; if a published map breaks that rule the values are reported as read and the violation is named.

Independent check. Every job reads each band once more at the coordinate itself, separately from the grid. The grid value at the site cell must equal that read; a job that fails the check is refused, because a shifted or transposed grid would look plausible and be wrong.

Layer status. Each layer reports ok (published data), empty_confirmed (the source was read in full and holds nothing in the box, the expected answer for the coastal layer at an inland site), unavailable (no coverage) or failed (not read; fetch again, and never read a failed layer as dry). A job is complete when both layers are ok or empty_confirmed, and partial when one layer carries data and the other does not.

The flood field has no climate scenarios or time horizons. It reports the published hazard maps at their reference period. For physical climate hazards under scenarios and horizons, including river flood and sea level rise ratings, see the Climate Risk methodology.

Site Terrain (CDEM)

The Site Terrain service resolves one coordinate into three aligned layers over a 2 km square: a bare-earth elevation grid, building footprints and road centrelines. It computes nothing; it delivers the ground that other services and your own models work on.

Bare earth. The elevation grid is GEDTM30 at its native 1 arc-second, about 30 m, with buildings and vegetation removed by the dataset’s own method, on the EGM2008 vertical datum. Cells are rectangular and grow narrower toward the poles because the grid is angular: about 30 m by 31 m near the equator and 19 m by 31 m at 52 N. Each cell carries a vertical uncertainty from the dataset; the median over the box is reported with every terrain and simulation result so depths can be judged against it. On flat urban ground that uncertainty is typically 0.4 to 0.7 m, which bounds what any depth read from this terrain can mean.

Buildings. Footprints come from one of two published sources, chosen by where the site is: Google Open Buildings where it covers, Microsoft building footprints everywhere else. Which source served the site, and why, is recorded in the response. Microsoft carries building heights where it has them; Google publishes none. Footprints are fetched over the box plus a 50 m margin so a building on the edge keeps its whole outline.

Roads. Centrelines with class, surface and bridge or tunnel flags, from the Overture Maps transportation theme built on OpenStreetMap. Roads describe access; they take no part in the simulation.

Layer status. Each layer reports ok, empty_confirmed, unavailable or failed, with the same meanings as the flood field. The distinction matters most for buildings: an empty building layer that is empty_confirmed is open land, and a failed building layer is a district that must not be modelled as bare earth. The terrain layer is the spine; without it there is no bundle.

What the simulation uses. The bare-earth grid, its vertical uncertainty and the building footprints. Roads are not used.

Pluvial and Flash Flood Simulation

Pluvial flooding is rain falling on the site and pooling or flowing over the ground before it can drain away. The Pluvial and Flash Flood service drops storms you choose on the site terrain and solves how the water moves.

The solver. A local-inertial shallow water solver after Bates, Horritt and Fewtrell (2010), with the flux form of de Almeida and others (2012) and building blocks after Schubert and Sanders (2012). The time step adapts to the flow. Rainfall is applied at a constant rate over the storm duration. The same solver runs the Pluvial Flood Simulator on the platform; the two are held byte-identical.

The mesh. The 30 m terrain is subdivided three times to a routing mesh of about 10 m by bilinear interpolation. This gives the water a finer grid to move on; it does not add terrain detail that the 30 m source does not hold.

Buildings. Every footprint is stamped into the mesh as a solid block so water flows around it and can be trapped in enclosed courtyards. Rain that lands on a roof is delivered to the nearest ground cell, the way a downspout does, so roofs read as dry and the ground around them carries the rain.

The box edge. The boundary is open: water that reaches the edge of the 2 km box leaves it, and the volume that left is reported. There is no drainage network inside the box.

The rainfall is yours. Each storm is a rainfall depth in millimetres over a duration in minutes, with a note naming the rainfall curve it came from. The service derives no return period from any curve, because a return period belongs to a gauge record and a published curve, and the choice of curve is part of the assessment. The note is echoed and not checked.

Two readings from one request. A one-hour storm is the pluvial reading: the depth of water in every cell at the two-hour mark, with the deepest water each cell saw beside it, shows where water stands after the rain and how deep it got on slopes and flow paths. A storm shorter than an hour is the flash flood reading: the end state understates what happened, so the reading adds time. The depth curve at the site every 30 seconds, the peak and the minute it was reached, and the first moment each cell reached 0.3 m and 0.5 m carry the signal. Both readings use the same physics, terrain and two-hour window, so they sit side by side.

The thresholds. 0.3 m and 0.5 m are the still-water depth limits of Australian Disaster Resilience Guideline 7-3 Flood Hazard (2017), Table 2: above 0.3 m flood water is unsafe for small vehicles, and above 0.5 m it is unsafe for vehicles, children and the elderly. Every cell records the first moment it crossed each one.

Outputs per storm.

• Depth in every cell at the stop time, and the deepest water in every cell.
• The time each cell first reached 0.3 m and 0.5 m.
• The depth curve every 30 seconds at the site, at the nearest ground cell when the site is on a roof, and at any probe you name.
• Statistics over the box, computed twice: over all cells and over land cells (ground above 0.5 m that is not a building).
• The volume stored and the volume that left.
• The solver’s mass balance, step count and build, carried as provenance.

Reading the numbers. When the site pin lands on a roof, the figures at the nearest ground cell are the site’s figures. In a box that touches the sea, the terrain carries no sea floor, so sea cells pond rain as a false flood; the land-cell statistics and the site are the readings to quote. Which depth is the headline is the reader’s decision: the end depth shows standing water, the peak shows the deepest moment, and a site in a hollow is best described by the level reached and the minutes to each threshold.

Verification

The solver is verified, and it is not field-validated.

Mass conservation. Every run keeps a ledger: water on the grid plus water that left across the edge must equal the rain that fell. The error on the verification scenarios is of the order of 1e-16, machine precision, and every result reports its own mass balance.

Analytic tests. Thirteen scenarios run on every change to the solver and are checked bit for bit against a stored reference. They include:

• a single basin that fills to a flat pond and matches the analytic volume;
• a tilted plane where ground above the wet extent must stay dry;
• two basins across a saddle with hand-computed capacities;
• mass gates under storm rainfall with closed and open edges;
• drainage over an edge notch;
• the footprint stamping, enclosed courtyards and the roof-rain redirect;
• a no-buildings regression.

Platform parity. The solver inside the API, the solver inside the platform’s Pluvial Flood Simulator and the solver behind this site’s free simulator are the same code, held byte-identical by a checksum. A result from the API and a run in the simulator on the same terrain and storm agree.

Reproducibility. Results from different computing environments agree in step count, stop time and wet area, and in depth to well under a millimetre; every result names the environment that produced it.

No field validation. The solver has not been tested against observed floods. Its depths are for relative comparison and screening.

Limitations

Flood field

• Depths come from global river and coastal models at 93 m and 930 m. A site-level decision needs site-level hydraulic modelling.
• The riverine map marks inland depressions as river flood in regions with no permanent rivers; the field lets a reader see this, and the service does not interpret it.
• A coastal cell on the shoreline is partly sea; its depth is inundation over the land in a cell about 1 km across, a screening figure.
• Riverine and coastal are independent reads. Compound flooding is not modelled.
• Both maps assume static land use.
• The flood field has no climate scenarios or horizons. Climate scenarios and horizons for flood-related hazards are in the Climate Risk methodology.

Site terrain

• Roads lag the live map by weeks to months and are pinned to a release.
• Building sources refresh on their own schedules, so counts can change between runs, and Google publishes no heights.
• The grid is bare earth; buildings are stamped only inside the simulation.

Pluvial and flash flood

• No infiltration: every millimetre of rain runs off, so every depth is an upper bound on ponding over permeable ground.
• No drainage network: no sewers, culverts or storm drains, so ponding is biased high in served urban catchments.
• The local-inertial form suits floodplain flow and is less suited to hydraulic jumps and fast, steep flow.
• Subdivision is not resolution: the 10 m mesh interpolates 30 m terrain.
• Terrain quality bounds everything: depressions shallower than the terrain’s 0.4 to 0.7 m uncertainty are inside its error.
• Pluvial only: rain inside the box, no river inflow, no channel routing from outside, no coastal boundary.
• No field validation.
• The box is 2 km across. Larger catchments are a different piece of work.
• Sea cells in a coastal box pond rain as a false flood; read the land-cell statistics and the site.
• The rainfall depth and its return period are the caller’s choice and are not checked.

Frequently Asked Questions

Why is the flood depth at my site null?
The site cell lies outside the modelled floodplain of the published river map. The response gives the distance to the nearest modelled cell so you can see how close the floodplain comes. A null is a statement about the map’s coverage, and the map holds no zero cells, so it is kept apart from “dry”.

Why are there no future scenarios in the flood field?
The flood field reports the published hazard maps at their reference period. Scaling a cell’s depth by a projected factor does not move water into the neighbouring cells where a deeper flood would spread, so a per-cell projection was retired. Scenario and horizon views of flood-related hazards are in the Climate Risk module.

Can I choose the rainfall for the simulation?
Yes, and you must. Each storm is a depth over a duration, with a note naming the curve it came from. Return periods belong to published rainfall curves for a gauge or region, and the choice is part of the assessment.

Why does the simulation show the deepest water in the sea?
Bare-earth terrain carries no sea floor, so in a coastal box the sea cells pond rain like a flat basin. The response reports statistics over land cells separately, and the site figures, so the false flood is visible and not quoted.

Is the simulation validated?
It is verified and not field-validated. Mass is conserved to machine precision and thirteen analytic scenarios pass on every change, but the solver has not been tested against observed floods. Use it for relative comparison and screening.

Is the simulator on this site the same as the API?
Yes. The same solver code runs in the free simulator, the platform module and the API, held byte-identical by a checksum.

Appropriate use: Portfolio screening and site comparison. Reading whether rivers or the sea reach a site, and how far the modelled floodplain is. Seeing where rain pools and flows across a site and how fast it arrives. Comparing storms, durations and sites on one method. Disclosure support, with the data sources, their licences and the method’s limitations cited alongside the figures.

Not suitable for: Flood insurance rating or underwriting, structural or drainage design, life-safety decisions, official flood zone determinations, real-time flood response, or any figure read as a measured depth at a door. For those, site-specific hydraulic modelling by qualified engineers is required.