Risk indicators — Global Forest Watch (live, needs a key)¶

The third risk source is the Global Forest Watch (GFW) Data API. It serves forest indicators — here annual tree-cover loss by country — and the GADM admin geometry those indicators are computed over.

Unlike ThinkHazard! and INFORM, GFW needs a free API key:

1. Create one by following the GFW guide: https://www.globalforestwatch.org/help/developers/guides/create-and-use-an-api-key/.
2. Make it available to the notebook via the GFW_API_KEY environment variable (e.g. export GFW_API_KEY=... before launching Jupyter, or set it in your shell profile).

The backend reads GFW_API_KEY from the environment automatically — the key is never passed in code or printed in this notebook. If the key is missing, constructing a gfw:* request raises an AuthenticationError naming GFW_API_KEY.

In [ ]:

import tempfile

import matplotlib.pyplot as plt

from earthlens.core import EarthLens

OUT = tempfile.mkdtemp()  # the tabular download also writes a CSV here

import tempfile import matplotlib.pyplot as plt from earthlens.core import EarthLens OUT = tempfile.mkdtemp() # the tabular download also writes a CSV here

Annual tree-cover loss by country¶

variables=["gfw:tree_cover_loss"] runs GFW's SQL query for the chosen country (canopy density ≥ 30%) and returns one row per year. The data_source="gfw" alias reaches the same risk-indicators backend.

In [ ]:

loss = EarthLens(
    data_source="gfw",
    variables=["gfw:tree_cover_loss"],
    country="KEN",
    path=OUT,
).download()

print(loss.shape)
loss.head()

loss = EarthLens( data_source="gfw", variables=["gfw:tree_cover_loss"], country="KEN", path=OUT, ).download() print(loss.shape) loss.head()

umd_tree_cover_loss__year is the year and umd_tree_cover_loss__ha the hectares of tree-cover loss in that year (UMD/Hansen). Plain pandas from here — plot the annual series:

In [ ]:

fig, ax = plt.subplots(figsize=(11, 4))
ax.bar(
    loss["umd_tree_cover_loss__year"],
    loss["umd_tree_cover_loss__ha"],
    color="tab:green",
    alpha=0.8,
)
ax.set_title("Kenya — annual tree-cover loss (canopy >= 30%)")
ax.set_xlabel("year")
ax.set_ylabel("tree-cover loss (ha)")
fig.tight_layout()

fig, ax = plt.subplots(figsize=(11, 4)) ax.bar( loss["umd_tree_cover_loss__year"], loss["umd_tree_cover_loss__ha"], color="tab:green", alpha=0.8, ) ax.set_title("Kenya — annual tree-cover loss (canopy >= 30%)") ax.set_xlabel("year") ax.set_ylabel("tree-cover loss (ha)") fig.tight_layout()

The admin boundary as a FeatureCollection¶

gfw:admin_boundary is a vector dataset: it returns the GADM country geometry the indicators are computed over as a pyramids FeatureCollection (a GeoDataFrame subclass), not a table.

In [ ]:

boundary = EarthLens(
    data_source="gfw",
    variables=["gfw:admin_boundary"],
    country="KEN",
    path=OUT,
).download()

print("type:", type(boundary).__name__)
print("features:", len(boundary))
print("crs:", boundary.crs)
print("bounds:", boundary.total_bounds)

boundary = EarthLens( data_source="gfw", variables=["gfw:admin_boundary"], country="KEN", path=OUT, ).download() print("type:", type(boundary).__name__) print("features:", len(boundary)) print("crs:", boundary.crs) print("bounds:", boundary.total_bounds)

It carries a CRS (EPSG:4326) and a geometry column, so it plots directly:

In [ ]:

ax = boundary.plot(facecolor="none", edgecolor="tab:green", linewidth=1.2)
ax.set_title("Kenya — GADM admin boundary (GFW geostore)")
ax.set_xlabel("longitude")
ax.set_ylabel("latitude")

ax = boundary.plot(facecolor="none", edgecolor="tab:green", linewidth=1.2) ax.set_title("Kenya — GADM admin boundary (GFW geostore)") ax.set_xlabel("longitude") ax.set_ylabel("latitude")

Takeaway¶

• GFW needs a free key in GFW_API_KEY; the backend reads it from the environment — never hard-code or print it.
• gfw:tree_cover_loss is tabular (a per-year DataFrame); gfw:admin_boundary is vector (a FeatureCollection that plots directly) — the per-instance OUTPUT_KIND in action.
• The same EarthLens(...).download() call shape serves all three risk sources; see the quickstart for the public ThinkHazard! / INFORM sources and catalog & behaviour for the full dataset list.