DWD RADKLIM / RADOLAN — catalog explorer (no network)¶

Explore the earthlens.radklim backend offline: the four products, how a request enumerates the raw granules it would download, and a real RADOLAN granule read + plotted with pyramids. Everything here runs deterministically at docs-build time — no DWD access.

RADKLIM is DWD's gauge-adjusted radar precipitation over Germany: the reprocessed climatology RADKLIM (statistics) and the operational near-real-time RADOLAN stream. See the API reference.

Setup¶

Imports up front. earthlens.radklim.Catalog loads the bundled product catalog (no network); the EarthLens facade builds a request; pyramids reads the granule; pandas / matplotlib render the tables and figure. DATA is a notebook-relative path to the docs example-data folder.

In [ ]:

import tempfile
from pathlib import Path

import numpy as np
import pandas as pd
from pyramids.dataset import Dataset

from earthlens.core import EarthLens
from earthlens.radklim import Catalog
from earthlens.radklim._helpers import operational_granule_url

DATA = Path('../../../examples/data/radklim')

import tempfile from pathlib import Path import numpy as np import pandas as pd from pyramids.dataset import Dataset from earthlens.core import EarthLens from earthlens.radklim import Catalog from earthlens.radklim._helpers import operational_granule_url DATA = Path('../../../examples/data/radklim')

The product catalog¶

Instantiate the catalog and read its licence and grid. RADKLIM is DWD open geodata under CC-BY-4.0 / GeoNutzV — attribution to Deutscher Wetterdienst (DWD) is required.

In [ ]:

cat = Catalog()
print('products :', cat.products())
print('license  :', cat.license)
print('grid     :', cat.grid['id'])

cat = Catalog() print('products :', cat.products()) print('license :', cat.license) print('grid :', cat.grid['id'])

Products at a glance¶

The four products split across two streams. reproc (RADKLIM) is the climatology — one yearly NetCDF archive per year, full 2001- record, use it for statistics. operational (RADOLAN) is near-real-time — per-timestamp granules on a rolling ~2-day window. Each comes in an hourly (RW) and a 5-min (YW) flavour.

In [ ]:

rows = [
    {
        'product': k,
        'stream': p.stream,
        'cadence': p.cadence,
        'format': p.default_format,
        'retention_days': p.retention_days or '-',
    }
    for k, p in cat.datasets.items()
]
pd.DataFrame(rows).set_index('product')

rows = [ { 'product': k, 'stream': p.stream, 'cadence': p.cadence, 'format': p.default_format, 'retention_days': p.retention_days or '-', } for k, p in cat.datasets.items() ] pd.DataFrame(rows).set_index('product')

RADKLIM-YW (5-min, 1 km) is the single most useful dataset for German sub-hourly extreme rainfall. The record is ~25 years — excellent for event structure, short for long return periods. For return-period work prefer RADKLIM over the inhomogeneous operational RADOLAN stream.

The download plan (reproc)¶

Build a request through the EarthLens facade with dataset='radklim-yw'. The reprocessing has no finer addressable unit than the year, so a [start, end] window maps to one yearly .tar.gz NetCDF archive per year. _search() returns that plan without touching the network — a cheap dry-run of what download() would fetch.

In [ ]:

lens = EarthLens(
    data_source='radklim',
    dataset='radklim-yw',
    start='2020-06-01',
    end='2022-06-01',
    lat_lim=[47.0, 55.0],
    lon_lim=[6.0, 15.0],
    path=tempfile.mkdtemp(prefix='radklim_'),
)
plan = lens.datasource._search()
pd.DataFrame([{'id': p.id, 'file': p.href.rsplit('/', 1)[-1]} for p in plan])

lens = EarthLens( data_source='radklim', dataset='radklim-yw', start='2020-06-01', end='2022-06-01', lat_lim=[47.0, 55.0], lon_lim=[6.0, 15.0], path=tempfile.mkdtemp(prefix='radklim_'), ) plan = lens.datasource._search() pd.DataFrame([{'id': p.id, 'file': p.href.rsplit('/', 1)[-1]} for p in plan])

Three years in the window → three yearly archives. These are large (YW ~13.5 GB/yr, RW ~836 MB/yr), so download() streams them to disk — this notebook only inspects the plan.

Operational granule URLs¶

The operational stream is addressed per timestamp instead. A granule name carries a YYMMDDHHMM stamp; the helper builds its URL. earthlens reads the stream's directory listing and keeps the granules inside the request window (behind the ~2-day retention guard).

In [ ]:

operational_granule_url('yw', 'raa01-yw_10000-2608101820-dwd---bin.hdf5')

operational_granule_url('yw', 'raa01-yw_10000-2608101820-dwd---bin.hdf5')

Read a real RADOLAN granule with pyramids¶

earthlens returns raw granule paths; reading them is pyramids' job. A small real operational RADOLAN-RW HDF5 granule ships with the docs. It opens directly on the fixed RADOLAN polar-stereographic grid over Germany.

In [ ]:

granule = DATA / 'raa01-rw_10000-2608101830-dwd---bin.hdf5'
ds = Dataset.read_file(granule)
print('shape:', ds.shape)
print('crs  :', ds.crs.split('PROJECTION')[-1][:40], '...')

granule = DATA / 'raa01-rw_10000-2608101830-dwd---bin.hdf5' ds = Dataset.read_file(granule) print('shape:', ds.shape) print('crs :', ds.crs.split('PROJECTION')[-1][:40], '...')

Plot the precipitation field¶

Read the array and mask the negative no-data cells, then show the hourly precipitation over the RADOLAN grid. The Germany outline of the composite is clearly visible.

In [ ]:

arr = ds.read_array()
band = arr[0] if getattr(arr, 'ndim', 2) == 3 else arr

# Plot the granule itself rather than rebuilding it through from_array.
# RADOLAN's grid is a DWD polar-stereographic PROJCS with no EPSG code, so
# `ds.epsg` is None and GeoReference — which carries only an EPSG — would drop
# the projection silently. The old `band < 0` guard is gone too: the band is
# uint16, so it could never fire.
glyph = ds.plot(
    cmap='Blues',
    vmax=float(np.percentile(band, 99)),
    title='RADOLAN-RW operational granule — (c) Deutscher Wetterdienst (DWD)',
)
glyph.cbar.set_label('hourly precipitation (raw units)')

arr = ds.read_array() band = arr[0] if getattr(arr, 'ndim', 2) == 3 else arr # Plot the granule itself rather than rebuilding it through from_array. # RADOLAN's grid is a DWD polar-stereographic PROJCS with no EPSG code, so # `ds.epsg` is None and GeoReference — which carries only an EPSG — would drop # the projection silently. The old `band < 0` guard is gone too: the band is # uint16, so it could never fire. glyph = ds.plot( cmap='Blues', vmax=float(np.percentile(band, 99)), title='RADOLAN-RW operational granule — (c) Deutscher Wetterdienst (DWD)', ) glyph.cbar.set_label('hourly precipitation (raw units)')

Takeaway¶

• Four products, two streams: RADKLIM (reproc, yearly NetCDF archives, statistics) and RADOLAN (operational, per-timestamp HDF5, near-real-time).
• A request enumerates raw granules — yearly archives for reproc, in-window timestamps for operational — and download() returns their paths.
• Reading (NetCDF / HDF5) is pyramids; earthlens never imports wradlib / xarray / netCDF4.
• The data is DWD open geodata (CC-BY-4.0 / GeoNutzV) — always credit Deutscher Wetterdienst (DWD).