GIRI flood depth RP100 (baseline, 90 m)

Source: GIRI — Global Infrastructure Resilience Index riverine flood hazard (UNEP-GRID / CDRI, 2023), CC BY 3.0 IGO Year: Present climate (W5E5 forcing) Category: Physical risks · Direct Physical Loss engine inputs Coverage: Global Format: Raster grid (92.8 m) Used in risk analysis: Yes — the riverine arm of the Direct Physical Loss flood figure, and gates Climate hazards (Flood mitigation services) at ≥ 1.55 m since 2026-09-11

What it shows

How deep the water stands, in metres, when a once-in-a-century river flood covers the ground. "Once in a century" is the return period — the average interval between two floods at least that severe.

The depth is the flood itself, not what it costs, and it counts nothing that would hold the water back: this is the undefended hazard. Dikes and levees enter later, as a truncation of the annualised loss at the design standard they protect to, not as a lower depth.

Darwin imports the whole return-period set — 5, 25, 50, 100, 200, 500 and 1000 years — because an annualised loss integrates over all of them. This page documents the 100-year band, the operating point the app reports; the rest of the bundle stays hidden.

How it is built

GIRI runs a continuum hydrological model forced with the W5E5 meteorological dataset for a present-climate run, and converts simulated river discharge into inundation depth. The output is published at 3 arc-seconds — roughly 90 m at the equator — and Darwin keeps it there, resampling nearest-neighbour onto a 92.766 m web-mercator grid. The centimetre-to-metre conversion is baked at production time.

How to read it

Read it as a depth in metres at a fixed rarity: the one-in-a-hundred-year flood. Below 0.5 m is very low, 0.5–1 m low, 1–2 m moderate, 2–4 m high, above 4 m very high — the same bands the coastal depth layer uses, because "how deep would the water be" is comparable across sources while the return period is not.

Dry and no-data look alike, and both are handled. The raster carries two no-data sentinels rather than one: the warp writes −9999 where GIRI reports no flooding, and the storage format keeps a wholly-unflooded tile as a hole that reads back as 0. Declaring only one of the two would paint the entire dry world in the colour of 0 m of water.

Class thresholds

A site is flagged on Climate hazards (Flood mitigation services) when the 100-year depth reaches 1.55 m, and the flag is very material when the entity's “Flood mitigation services” pressure is also ≥ 4. The colour ramp on the map is continuous, from 0 to 6 m; the bands quoted under How to read it are a reading aid, not classes the layer publishes.

How this threshold was calibrated

The flag used to read Flood depth — 10-year return, present climate (GIRI) at ≥ 1 m. Moving it here changed both the return period and the unit — a 10-year depth in centimetres to a 100-year depth in metres — so no number could be carried over; the threshold was measured. On the two rasters, the surface the old rule flagged (10-year depth ≥ 1 m) is reproduced by this layer at ≥ 1.55 m. A round 1.50 m would flag ×1.02 that surface, inside the noise of the method, so 1.55 m is kept as measured.

The measure is the flagged surface, not a percentile of depth. Taking the same percentile over each raster's own wet extent gives 0.85 m instead, and it is the wrong comparison: the 100-year flood spreads further, and the extra fringe is shallow, which drags the percentile down. The flagged surface is what a reader sees on the map, and it compares across two different grids.

An open question travels with it, in the new unit: the lower target discussed for the old rule, a 10-year depth of 50 cm, corresponds to 1.00 m here on the same measurement. That choice is still to be arbitrated.

Why this layer

Both arms of a flood figure now read at the same resolution. A flood loss sums a riverine arm and a coastal arm. The coastal arm has read Deltares coastal flood depth RP100 at 92.8 m since August 2026; the riverine arm was still being sampled on a ~2 km exchange resample of the same GIRI data — a twentyfold gap inside a single number, on the hazard whose loss is the most sensitive of the set to the exact coordinate. At 2 km, a point in a city centre reading metres of water and a business park a kilometre away reading dry sit inside one pixel. This layer is that arm rebuilt on GIRI's native grid, matching the coastal set to the pixel.

The 2 km bundle stays registered but hidden, and its name now says what it is. It remains the grid against which the Aqueduct change factors were validated, and nothing in the engine reads it any more.

The engine reads it over the site, not under its pin. Since 2026-09-15 a site drawn as a polygon samples this grid across its whole shape and keeps, per return period, the depth reached by its wettest 20 % of points; every site also reads a 500 m ring around it, which prices the disruption of a dry site in a flooded district. The map shows the raster itself; the damage functions page describes both readings.

Aqueduct deliberately stays at 2 km. The forward-looking flood signal comes from WRI Aqueduct as a dimensionless ratio per return period, never an absolute depth — so its resolution cancels in the ratio and buys nothing by being refined.

Source

GIRI — Global Infrastructure Resilience Index riverine flood hazard (UNEP-GRID / CDRI, 2023), CC BY 3.0 IGO. https://giri.unepgrid.ch/. Continuum hydrological model forced with W5E5, present-climate run.

Comparison with the WWF Risk Filter Suite

WWF's flood hazard component is JRC fluvial RP100, classified into relative quantiles across basins. This layer is an absolute depth at the same return period, on a finer grid and from a different model chain, read at the site rather than at its basin. Since 2026-09-11 this layer also carries the riverine screening flag; the 10-year GIRI view, Flood depth — 10-year return, present climate (GIRI), stays in the catalogue as a display layer with its own class bands.