HANZE — floods and their impacts, in five minutes¶

HANZE (Historical Analysis of Natural Hazards in Europe) is the observed record of European flood events and their impacts — real floods since 1870, each with its fatalities, persons affected, area flooded and economic losses. This is the observed hazard → loss side of the flood chain: the yardstick a modelled event set is validated against.

This notebook pulls German and Dutch flood events straight from HANZE's pinned Zenodo release (CC-BY-4.0, no credentials), looks at the loss and fatality distribution, and maps the affected NUTS-3 regions. Everything runs through the unified EarthLens facade.

In [ ]:

import tempfile
from pathlib import Path

import matplotlib.pyplot as plt
import pandas as pd

from earthlens.core import EarthLens

# A stable scratch directory keeps the small (~3 MB) source files cached between runs.
OUT = Path(tempfile.gettempdir()) / "earthlens-hanze-demo"
OUT.mkdir(exist_ok=True)

import tempfile from pathlib import Path import matplotlib.pyplot as plt import pandas as pd from earthlens.core import EarthLens # A stable scratch directory keeps the small (~3 MB) source files cached between runs. OUT = Path(tempfile.gettempdir()) / "earthlens-hanze-demo" OUT.mkdir(exist_ok=True)

1. Pull the events¶

hanze is a tabular backend: download() returns a pandas.DataFrame, one row per historical flood, with HANZE's own documented columns. We select Germany and the Netherlands over 1950–2020 with the country= / start= / end= facets — no variables needed.

In [ ]:

events = EarthLens(
    "hanze",
    start="1950",
    end="2020",
    country=["DE", "NL"],
    path=str(OUT),
).download(progress_bar=False)

print(f"{len(events)} flood events for DE + NL, 1950-2020")
events[
    [
        "Country code",
        "Year",
        "Type",
        "Fatalities",
        "Persons affected",
        "Losses (real value)",
    ]
].head(10)

events = EarthLens( "hanze", start="1950", end="2020", country=["DE", "NL"], path=str(OUT), ).download(progress_bar=False) print(f"{len(events)} flood events for DE + NL, 1950-2020") events[ [ "Country code", "Year", "Type", "Fatalities", "Persons affected", "Losses (real value)", ] ].head(10)

2. What kind of floods, and how damaging?¶

The Type column is one of River, Flash, Coastal, or River/Coastal. Let's count the events by type, then look at the distribution of inflation-adjusted (2025 €) losses. Losses are reported values, not exposure-normalised — see the backend's introduction page for why that distinction matters.

In [ ]:

losses = pd.to_numeric(events["Losses (real value)"], errors="coerce")
fatalities = pd.to_numeric(events["Fatalities"], errors="coerce")

fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(12, 4))

events["Type"].value_counts().plot.bar(ax=ax1, color="#4C72B0")
ax1.set_title("Flood events by type (DE + NL, 1950-2020)")
ax1.set_xlabel("Type")
ax1.set_ylabel("Number of events")
ax1.tick_params(axis="x", rotation=0)

valid = losses[losses > 0]
ax2.hist(valid, bins=20, color="#C44E52")
ax2.set_title("Distribution of real losses")
ax2.set_xlabel("Losses (real value, 2025 EUR)")
ax2.set_ylabel("Number of events")

fig.tight_layout()
plt.show()

print(f"Total recorded fatalities: {int(fatalities.sum()):,}")
print(f"Deadliest single event: {int(fatalities.max()):,} fatalities")

losses = pd.to_numeric(events["Losses (real value)"], errors="coerce") fatalities = pd.to_numeric(events["Fatalities"], errors="coerce") fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(12, 4)) events["Type"].value_counts().plot.bar(ax=ax1, color="#4C72B0") ax1.set_title("Flood events by type (DE + NL, 1950-2020)") ax1.set_xlabel("Type") ax1.set_ylabel("Number of events") ax1.tick_params(axis="x", rotation=0) valid = losses[losses > 0] ax2.hist(valid, bins=20, color="#C44E52") ax2.set_title("Distribution of real losses") ax2.set_xlabel("Losses (real value, 2025 EUR)") ax2.set_ylabel("Number of events") fig.tight_layout() plt.show() print(f"Total recorded fatalities: {int(fatalities.sum()):,}") print(f"Deadliest single event: {int(fatalities.max()):,} fatalities")

3. Where did they happen?¶

Pass with_geometry=True and download() returns a pyramids FeatureCollection of the affected NUTS-3 regions instead of the events table. Each event's Regions affected (NUTS 3) list is joined to the NUTS-3 boundary polygons; every affected region comes back once with an n_events count — a ready-made choropleth. The polygons are reprojected from the release's ETRS89-LAEA (EPSG:3035) to WGS84 for you.

In [ ]:

regions = EarthLens(
    "hanze",
    start="1950",
    end="2020",
    country=["DE", "NL"],
    with_geometry=True,
    path=str(OUT),
).download(progress_bar=False)

print(f"{len(regions)} affected NUTS-3 regions; CRS EPSG:{regions.crs.to_epsg()}")
regions[["nuts3_code", "region_name", "n_events"]].sort_values(
    "n_events", ascending=False
).head(10)

regions = EarthLens( "hanze", start="1950", end="2020", country=["DE", "NL"], with_geometry=True, path=str(OUT), ).download(progress_bar=False) print(f"{len(regions)} affected NUTS-3 regions; CRS EPSG:{regions.crs.to_epsg()}") regions[["nuts3_code", "region_name", "n_events"]].sort_values( "n_events", ascending=False ).head(10)

In [ ]:

ax = regions.plot(
    column="n_events",
    cmap="OrRd",
    legend=True,
    edgecolor="0.6",
    linewidth=0.3,
    figsize=(8, 9),
    legend_kwds={"label": "Flood events, 1950-2020", "shrink": 0.6},
)
ax.set_title("Affected NUTS-3 regions by flood-event count (DE + NL)")
ax.set_axis_off()
plt.show()

ax = regions.plot( column="n_events", cmap="OrRd", legend=True, edgecolor="0.6", linewidth=0.3, figsize=(8, 9), legend_kwds={"label": "Flood events, 1950-2020", "shrink": 0.6}, ) ax.set_title("Affected NUTS-3 regions by flood-event count (DE + NL)") ax.set_axis_off() plt.show()

Recap¶

• EarthLens("hanze", country=[...], start=, end=) → a DataFrame of observed floods and their impacts.
• with_geometry=True → a FeatureCollection of the affected NUTS-3 regions, one per region with an n_events count, in WGS84.
• The data is HANZE v3.0.1-beta (pinned Zenodo record 20478847, CC-BY-4.0) — cite Paprotny et al., and treat the figures as provisional.
• Losses are reported (nominal + inflation-adjusted real), not exposure-normalised.