Skip to content

Get started

Install EuroFlood, run your first flood query, and download a depth map, end to end, against the published index (no data build required). Prefer runnable notebooks? The Tutorials cover all of this in depth.

Install

pip install "euroflood[viz]"   # core + plotting (recommended)

uv add "euroflood[viz]"

EuroFlood works out of the box: the published index is read remotely and cached on first use (~14 MB of tables), so there is nothing to download or configure first.

Your first query

floods(...) returns a FloodFrame: a geopandas.GeoDataFrame, one row per historic flood event. It's cheap: it streams a small window of the index and downloads no rasters. That's the discover step of EuroFlood's discover → extract model; you extract depth rasters only for the events you keep.

import euroflood as ef

cat = ef.floods("Zutphen, Netherlands")
cat

EuroFlood catalogue: 25 flood events · 2015-01-12 … 2024-02-05 · 51.6 km² total

Because it is a GeoDataFrame, filter and plot it as usual, then .download() the depth rasters for the events you keep and .stats() them (no network after the download):

recent = cat[cat["date"] >= "2021-01-01"]    # any pandas / geopandas operation
recent.plot()                                 # recurrence heatmap, still no download
dl = recent.download("out/")                  # fetch + crop only these events' rasters
dl.stats()                                    # max/mean/p95 depth (m), area (km²), volume

Ways to specify a region

A place name is geocoded; a bbox, point + radius_m, or shapefile skip geocoding. Administrative boundaries can be awkward (they often follow a river), so shape="bbox" / "hull" gives a cleaner ROI, and buffer_m grows it:

ef.floods("Zutphen, Netherlands", buffer_m=1000)
ef.floods("Zutphen, Netherlands", shape="bbox")   # a clean bounding-box ROI
ef.floods(bbox=(6.14, 52.09, 6.27, 52.17))
ef.floods(point=(52.14, 6.20), radius_m=6000)     # point is (lat, lon)

Modelled hazard

The same API queries the global CEMS-GLOFAS flood-hazard maps by return period (seven are available: 10, 20, 50, 75, 100, 200, and 500 years):

haz = ef.hazard("Zutphen, Netherlands", return_period=[100, 500]).download("hazard/")
haz.stats()

Offline & HPC

Everything above streams data on demand. For a fully offline / cluster node, mirror the layers you need once (on a machine with internet), then flip the node offline:

# On a networked login node, stage a study region for offline use:
euroflood mirror all --bbox 6.1 52.0 6.3 52.2 -r 100   # index + flood depths + hazard tiles
euroflood verify all --bbox 6.1 52.0 6.3 52.2 -r 100 --deep   # readiness gate (checksums)

# On the offline compute node, one switch forces everything cache-only:
export EUROFLOOD_OFFLINE=1
euroflood floods --bbox 6.1 52.0 6.3 52.2 --download --out out/
euroflood hazard --bbox 6.1 52.0 6.3 52.2 -r 100 --download --out out/

mirror stages any layer independently: mirror index (the flood catalogue, so floods() queries run offline), mirror floods --bbox … (the flood depth maps for a region), mirror hazard --bbox … (GLOFAS hazard tiles for a region), or mirror all for everything. verify reports what is present / missing / corrupt (--deep re-checks sha256). EUROFLOOD_OFFLINE=1 (or euroflood.offline() in Python) forces both collections cache-only and the geocoder to the offline NUTS backend; a missing tile then raises a clear error naming the exact mirror command to run, never a silent partial result. In Python: ef.mirror("hazard", bbox=(6.1, 52.0, 6.3, 52.2), return_period=100).

Next steps

  • Tutorials

    The guided path: from a first query to hazard maps and quantitative analysis.

    Start the tutorials

  • Case studies

    Real-world flood analyses on genuine events across Europe, from Storm Boris to the Valencia DANA.

    Browse case studies

  • Concepts

    How the index works: the one page that makes everything else click.

    Read Concepts

  • API reference

    Every public function, class, and EUROFLOOD_* setting.

    Browse the API