DWD ICON precipitation — direct-HTTPS fetch¶

DWD Open Data publishes ICON forecasts as per-variable, bz2-compressed GRIB2 files over plain HTTPS (no SDK, no auth). earthlens' DWD centre downloads and decompresses them.

Requirements (live download):

pip install earthlens[nwp]

!!! note "Native grid" DWD's native ICON-global files are on an icosahedral grid, which is not a regular lat/lon raster — so this notebook fetches the raw GRIB2 rather than cropping it to a COG. For a croppable COG use a regular-lat/lon ICON product (e.g. icon-eu).

Setup¶

The imports for the notebook. earthlens.nwp provides the Catalog of forecast models and the DWDCentre that fetches DWD Open Data GRIB2 files directly over HTTPS.

In [ ]:

import datetime as dt
from pathlib import Path

from earthlens.nwp import Catalog
from earthlens.nwp.centres.dwd import DWDCentre

import datetime as dt from pathlib import Path from earthlens.nwp import Catalog from earthlens.nwp.centres.dwd import DWDCentre

1 · The ICON-global model¶

Look up the icon-global model in the catalog and inspect its band table. Each entry maps an earthlens variable name to the DWD GRIB short name (e.g. precipitation_acc → TOT_PREC).

In [ ]:

model = Catalog().get_model('icon-global')
model.bands

model = Catalog().get_model('icon-global') model.bands

2 · Pick a published forecast cycle¶

ICON-global runs at 00/06/12/18Z and DWD keeps only roughly the last day online, so we pick the most recent run that is already published: back off about 7 hours and floor to the 6-hourly cycle grid.

In [ ]:

# ICON-global runs 00/06/12/18Z and DWD keeps only ~the last day online,
# so pick the most recent run that is already published (~back off 7 h,
# floored to the 6-hourly cycle grid).
now = dt.datetime.now(dt.UTC).replace(tzinfo=None)
ref = now - dt.timedelta(hours=7)
cycle = ref.replace(hour=(ref.hour // 6) * 6, minute=0, second=0, microsecond=0)
cycle

# ICON-global runs 00/06/12/18Z and DWD keeps only ~the last day online, # so pick the most recent run that is already published (~back off 7 h, # floored to the 6-hourly cycle grid). now = dt.datetime.now(dt.UTC).replace(tzinfo=None) ref = now - dt.timedelta(hours=7) cycle = ref.replace(hour=(ref.hour // 6) * 6, minute=0, second=0, microsecond=0) cycle

3 · Fetch the precipitation GRIB2¶

Create the output directory, then ask the DWDCentre to fetch a single GRIB2 file: the precipitation_acc band of analysis step 0 for the chosen cycle. mirror='auto' selects a working DWD mirror; the call returns the path to the decompressed .grib2.

In [ ]:

out_dir = Path('out/icon')
out_dir.mkdir(parents=True, exist_ok=True)

out_dir = Path('out/icon') out_dir.mkdir(parents=True, exist_ok=True)

Run the fetch and inspect the resulting file's name and on-disk size.

In [ ]:

centre = DWDCentre(out_dir)
grib_path = centre.fetch_one(
    model, cycle, step=0, params=['precipitation_acc'], mirror='auto'
)
grib_path.name, grib_path.stat().st_size

centre = DWDCentre(out_dir) grib_path = centre.fetch_one( model, cycle, step=0, params=['precipitation_acc'], mirror='auto' ) grib_path.name, grib_path.stat().st_size

Recap¶

The downloaded .grib2 holds the requested band's decompressed messages. For a regular-grid model you would instead call EarthLens(data_source="nwp", variables={...}).download() and receive a bbox-cropped COG, exactly like the GFS quickstart.