Skip to content

13 A flood map for your town (2D)

Goal: a 2D flood simulation for one town, from satellite rain, with only the town's coordinates and the dates of the storm; then how big to make the area and how long it takes, from five places.

You need: an Earth Engine sign-in and project (setup) and pip install "MRRpy[gee]". In Jupyter, the same steps are in notebooks/flood_maps_2d.ipynb.

Settings used: INUNDATION_MAP, INUNDATION_METHOD, INUNDATION_AREA, INUNDATION_DEM, PRECIP_METHOD, EVENT_START_UTC, RUNOFF_SOURCE, MANNINGS_N_SOURCE, DEM_BOUNDS_WGS84, OUTPUT_POINT.

The example is Montpelier, Vermont, in the flood of 10–11 July 2023.

The steps

  1. The basin. A box around the river basin, for the terrain download, and the outlet: a point on the river at or just below the town. MRRpy finds the basin above it.
  2. The storm. The dates, for satellite rain (NASA IMERG), or rain gauges, or a design storm.
  3. The area for the flood map. A box around the town (see how big below).
  4. Run. MRRpy routes the river flow over the whole basin at 90 m, then simulates the flood in 2D over the box, on FABDEM at 30 m, with the routed rivers flowing in.
  5. Read the maps, and the warnings in the log: if the flood reached an edge of the box, make the box larger there and redraw.

Set it up

  1. Terrain: Download a DEM for an area, fabdem, Download resolution 90. Press Pick the outlet or download area on a map: click the river just below the town for the outlet, then, with A rectangle sets: Area to download, draw a rectangle around the basin with the square tool.
  2. Coordinate system: Use the outlet's UTM zone.
  3. Precipitation: IMERG satellite rain, event start 2023-07-09 00:00, simulation length 96 hours.
  4. Runoff: scs_cn.
  5. Routing: diffusive_implicit; roughness from land cover (lulc).
  6. Outputs and run: tick Flood depth and extent maps, How to spread the water 2d. Back on the map, choose A rectangle sets: Area for the flood maps and draw the box around the town. Then Run the model.
DEM_PATH: ''
DEM_SOURCE: fabdem
DEM_SCALE_M: 90.0
DEM_BOUNDS_WGS84: [-72.85, 43.95, -72.15, 44.62]
OUTPUT_POINT: [44.27504, -72.65924]
GEE_PROJECT: ee-yourname
OUTPUT_DIR: results/montpelier/
PRECIP_METHOD: imerg_thiessen
EVENT_START_UTC: '2023-07-09 00:00'
TOTAL_SIMULATION_TIME_HOURS: 96.0
RUNOFF_SOURCE: scs_cn
MANNINGS_N_SOURCE: lulc
ROUTING_SCHEME: diffusive_implicit
INUNDATION_MAP: true
INUNDATION_METHOD: 2d
INUNDATION_AREA: [-72.6527, 44.2201, -72.5122, 44.2847]

MRRpy run -c montpelier.yaml
# another box, your own LiDAR or the other method, without routing again:
MRRpy run -c montpelier.yaml --stages inundation

from MRRpy import Config, run_pipeline, plot_inundation, animate_inundation

cfg = Config(
    DEM_PATH="", DEM_SOURCE="fabdem", DEM_SCALE_M=90,
    DEM_BOUNDS_WGS84=(-72.85, 43.95, -72.15, 44.62),        # the basin
    OUTPUT_POINT=(44.27504, -72.65924),                     # the outlet, below the town
    GEE_PROJECT="ee-yourname", OUTPUT_DIR="results/montpelier/",
    PRECIP_METHOD="imerg_thiessen",
    EVENT_START_UTC="2023-07-09 00:00", TOTAL_SIMULATION_TIME_HOURS=96,
    RUNOFF_SOURCE="scs_cn", MANNINGS_N_SOURCE="lulc",
    ROUTING_SCHEME="diffusive_implicit",
    INUNDATION_MAP=True, INUNDATION_METHOD="2d",
    INUNDATION_AREA=(-72.6527, 44.2201, -72.5122, 44.2847),  # the town
)
out = run_pipeline(cfg)
plot_inundation(out)                                         # deepest water
animate_inundation(out, out_path="results/montpelier/flood.gif")

cfg.INUNDATION_AREA = (-72.62, 44.24, -72.57, 44.27)         # just downtown,
out2 = run_pipeline(cfg, stages=["inundation"])              # no routing again
  1. Tab 1 · DEM & Watershed: download FABDEM at 90 m for the basin box and delineate the watershed at the outlet.
  2. Tab 2 · Precipitation: IMERG from 2023-07-09 00:00 UTC. Tab 3 · Runoff: SCS curve numbers.
  3. Tab 4 · Routing: diffusive_implicit, 96 h; in Flood Depth and Extent Maps tick Make flood depth and extent maps, choose Flood simulation over one area (2D) and type the box in Map only this area: -72.6527, 44.2201, -72.5122, 44.2847.
  4. Run. On tab 5 · Results, Load Flood Maps adds the depth and the flooded area.

How big should the area be?

  • The river runs through the box, and its edges cross the valley rather than run along it.
  • Leave half a kilometre to a kilometre of valley beyond the town, upstream and downstream. Water leaves the box wherever the ground slopes out of it, but in a big flood it backs up towards the edge a river enters through. If the run says the flood reached an edge, extend the box there.
  • The outlet (OUTPUT_POINT) sits at or below the box, so every river in the box is routed.
  • Size: 20–100 km² at 30 m runs in seconds; up to about 500 km² is fine on a workstation. The limit is 5 million cells (about 4,500 km² at 30 m, 500 km² at 10 m).

Measured on a 32-core workstation, four days of storm, implicit routing at 90 m, the 2D on FABDEM at 30 m:

Place Basin Routing Area for the 2D 2D simulation
Ahr valley, Germany 854 km² (105,000 cells) 59 s 55 km² (61,000 cells) 9 s
Kathmandu, Nepal 606 km² (75,000 cells) 117 s 49 km² (55,000 cells) 20 s
Montpelier, Vermont 1,318 km² (163,000 cells) 139 s 87 km² (96,000 cells) 16 s
East Rapti, Chitwan 2,303 km² (284,000 cells) 655 s 172 km² (191,000 cells) 38 s
Pearl River at Jackson, Mississippi (a gauged inflow) – 520 km² (575,000 cells) 5 min for 15 days

The Earth Engine downloads (terrain, rain, curve numbers, land cover) add under a minute on a first run and are kept for reruns. On a laptop expect a few times longer.

Five places

Deepest water at Montpelier, Kathmandu, the East Rapti and the Ahr

The same settings at four places: FABDEM, IMERG rain, global curve numbers, land-cover roughness, no calibration.

Place Flood River flow, model vs gauge Flood map vs observations
Montpelier, Vermont July 2023 1,011 vs 654 m³/s all 40 USGS high-water marks inside the flood; levels 1.25 m high
Kathmandu, Nepal Sept 2024 1,875 vs about 1,020 m³/s (the record ends near the crest) 90 % of the flood traces seen by Sentinel-2
East Rapti, Chitwan Sept 2024 no gauge record at hand much wider than Sentinel-1 saw a day later
Ahr valley, Germany July 2021 gauge destroyed by the flood none used

What the places show:

  • The map follows the flows. Where gauges exist, satellite rain gave 1.5–1.8 times the peak flow, and the maps flood too much accordingly. Driven by the measured flows, the same 2D maps were within 0.4 m of the high-water marks on LiDAR and 1 m on FABDEM (6.3). With rain gauges or a gauged inflow, use them.
  • Flat valleys need large areas. In the flat Chitwan valley the flood spreads over kilometres; confined valleys hold it in a band.
  • A big river routed in from a gauge into a small model area (the Pearl, with Example 8's inflow points) travels too slowly: below an inflow point the channel is sized by the local drainage area. For a big river, route its whole basin.

Other options to try

Change Setting See
your own LiDAR for the flood map INUNDATION_DEM: file, INUNDATION_DEM_PATH: lidar.tif 6.3
quick maps for every river instead INUNDATION_METHOD: hand Example 12
rain gauges instead of satellite rain PRECIP_METHOD: thiessen Example 6
another box, without routing again MRRpy run -c montpelier.yaml --stages inundation 6.6