E4DRR/fastflood-ea-casestudy

FastFlood East Africa Flood Case Studies

FastFlood reproductions of the 11 ICPAC flood case studies, driven by the same event inputs as the RIM2D simulations in E4DRR/rim2d-simulations, and packaged in the same layout so the two methods compare map-for-map.

FastFlood (FastHazard, NL) is a fast steady-state flood method: DEM + rainfall (+ optional Manning / infiltration) → maximum inundation in seconds on CPU. RIM2D (GFZ Potsdam) is GPU 2D shallow-water hydrodynamics. This dataset lets you see how a lightweight screening method lines up against full 2D physics on identical events.

Produced under the ICPAC E4DRR / CRAF'D project.

Layout

Each <case>/ mirrors the RIM2D case layout:

<case>/
├── input/
│   ├── dem.tif              Copernicus GLO-30 elevation (m), stream-burned
│   ├── roughness.tif        Manning's n
│   ├── infiltration.tif     infiltration capacity (mm/hr)
│   ├── rain_mean_mmhr.tif   event-mean rainfall intensity (mm/hr)
│   └── permanent_water.tif  lake / reservoir / sea mask (1 = permanent water)
├── output/
│   ├── <case>_wd_max.tif    FastFlood maximum water-depth envelope
│   ├── frames/wh_*.tif      inundation at successive elapsed times
│   └── preview.gif          RIM2D-style water-depth animation
├── run_command.txt          the FastFlood invocation (≈ RIM2D .def)
└── METHOD.md                per-case method notes

Cases

Case	Country	Event	Duration
nairobi_2026-03-06	Kenya	flash flood, Nairobi metro	24 h
ssd_2025-10-05	South Sudan	Upper Nile / Renk	24 h
dji_2019-11-21	Djibouti	Djibouti City flash flood	72 h
eth_2021-08-16	Ethiopia	Addis Akaki river	72 h
som_2023-11-19	Somalia	Lower Shabelle	72 h
uga_2019-05-26	Uganda	central Uganda / Kampala	72 h
eri_2019-08-10	Eritrea	Asmara plateau highlands	120 h
rwa_2023-05-01	Rwanda	western + northern provinces	120 h
nile_2024_abu_hamad	Sudan	Abu Hamad (Nile)	144 h
tza_2024-04-10	Tanzania	Dar es Salaam metro	168 h
bdi_2024-04-01	Burundi	Bujumbura + central	504 h

Method

The RIM2D case NetCDFs (dem.nc, roughness.nc, pervious_surface.nc, and the per-30-min rain/imerg_t*.nc frames) are converted 1:1 to GeoTIFFs on the same 30 m UTM grid. FastFlood's -rain is an intensity applied as rain × dur, so the IMERG frames are reduced to an event-mean intensity map (total_mm = Σ(rate·0.5), mean = total_mm / duration_h).

Animations drive FastFlood with the cumulative IMERG hyetograph so inundation grows monotonically to the full-event result. Permanent water bodies (lakes, reservoirs, ocean) are masked (Natural Earth lakes + DEM sea) and shown as static dark water, not flood.

FastFlood vs RIM2D — flood-on-land extent

Wet area = depth > 5 cm; permanent water excluded from both models via the same per-case mask. RIM2D = case_v1_wd_max.nc reference.

Case	FF max (m)	FF wet %	RIM2D max (m)	RIM2D wet %
nairobi	10.53	6.2	20.08	11.8
ssd	0.97	4.2	5.24	5.7
dji	2.38	0.6	23.41	30.9
eth	2.22	1.8	21.45	3.9
som	1.43	3.1	7.57	37.0
uga	1.65	4.6	9.75	11.7
eri	1.55	0.8	12.91	3.6
rwa	3.02	5.2	16.78	11.7
nile	0.28	2.0	5.44	19.2
tza	2.54	4.0	11.58	32.6
bdi	2.14	1.5	16.93	27.7

FastFlood reproduces the same flood locations but under-predicts RIM2D. For inland cases the FastFlood/RIM2D wet-area ratio is a stable ~0.4–0.7; for coastal / lake cases (dji, som, tza, bdi) RIM2D floods extensive low coastal land that the steady pluvial FastFlood barely wets. This is expected for a fast steady-state approximation vs full 2D hydrodynamics — a screening tool, not a calibrated hazard product.

Reproduce

Pipeline code: https://github.com/nishadhka/DevOps-hazard-modeling/tree/main/fastflood (rim2d_nc_to_fastflood.py, run_all_cases.sh, make_fastflood_animation.py, build_water_mask.py, package_fastflood_case.sh).

License

CC BY 4.0. FastFlood is proprietary software of FastHazard (NL); this dataset contains only its inputs/outputs for these cases, not the software.