dwdopen

Python library to download numerical weather prediction (NWP) data from DWD Open Data. List currently available models, parameters, and runs. Select what data you want and download it as GRIB2 file(s).

Install

Requires Python 3.11 or newer.

pip install dwdopen

Quickstart

from dwdopen import DWD, Every

with DWD() as dwd:
    icon_eu = dwd.nwp.model("icon-eu")
    query = icon_eu.select(["T_2M", "TOT_PREC"], steps=Every("0h", "24h", "3h"))
    query.download("forecast.grib2")

This downloads 2 m temperature and total precipitation for the next 24 hours, every 3 hours, from the newest ICON-EU run that is completely published, into one GRIB2 file.

Discovery

dwd.nwp.models()                    # ['aicon', 'icon', 'icon-art', ..., 'icon-eu-eps']
icon_eu.parameters()                # ['ALB_RAD', 'ALHFL_S', ..., 'T_2M', ...]
icon_eu.parameter("T")              # on pressure (100) and model (150) levels
icon_eu.levels("T", "pressure")     # [50, 70, 100, ..., 850, ..., 1000]  (hPa)
icon_eu.runs()                      # [..., Run(2026-10-27T06:00Z)], oldest first

Examples

These continue inside the with DWD() as dwd: block of the quickstart.

Upper air: temperature on 850 and 500 hPa at +12 h.

upper_air = icon_eu.select("T", level_type="pressure", levels=[850, 500], steps=["12h"])
upper_air.download("t.grib2")

Ensemble: the first 5 members of ICON-EU-EPS, one file per member, into a directory.

from dwdopen import Between

eps = dwd.nwp.model("icon-eu-eps")
ensemble = eps.select("T_2M", members=Between(1, 5), steps="24h")
ensemble.download("members/", combine="member")

A scheduled job: give a directory instead of a file name, and the file is named after the model, run and selection. Running the job again for the same run downloads nothing. query is the one from the quickstart.

result = query.download("/data/icon-eu/")
# /data/icon-eu/icon-eu_2026-10-27T0600_TOT_PREC+T_2M_single_0h-24h_ec484355.grib2
result.assets_skipped   # 18 on the second run: the file was already there

Look before you download: resolve() freezes a query against one run and gives you a plan to inspect. The plan never switches to a newer run later.

plan = query.resolve()
print(plan)
# ResolvedRequest(run=2026-10-27T06:00Z, 18 assets, 12.1 MB, TOT_PREC, T_2M,
#                 9 steps 0h..24h)
plan.download("forecast.grib2")

The data is on the native ICON grid

DWD publishes every model on the unstructured triangular ICON grid. To get a regular lat-lon grid, regrid with CDO. For ICON global, DWD provides ready-made weights:

# from https://opendata.dwd.de/weather/lib/cdo/ICON_GLOBAL2WORLD_025_EASY.tar.bz2
cdo -f nc remap,target_grid_world_025.txt,weights_icogl2world_025.nc t_2m.grib2 t_2m.nc

For ICON-EU and the other models see Regridding in the guide.

Features

The guide explains each of these in detail.

• Every ICON model on DWD Open Data: ICON, ICON-EU, ICON-D2, ICON-D2-RUC, their ensembles, ICON-ART and AICON. Nothing is hard-coded.
• Steps as durations: "6h", ["0h", "3h"], Every("0h", "48h", "3h"), Between("0h", "48h"), "all", and 15 or 5 minute steps for ICON-D2 and ICON-D2-RUC.
• Vertical levels: pressure levels in hPa, model levels and soil levels.
• Ensemble members: members=[1, 2, 5], Between(1, 10), or all of them.
• ICON-ART wavelengths: wavelengths=532 for the dust backscatter fields.
• Complete runs only: by default a run that DWD is still publishing is skipped, and require="partial" takes what is there and reports the rest.
• Combining: return one file, one per parameter, one per ensemble member, or one file per GRIB message.
• Safe, resumable downloads: concurrent, written atomically, already downloaded files are skipped, with a progress bar.
• Clear errors: every error is a DWDOpenError, and typos in names and options are caught before anything is downloaded.

Roadmap

After 1.0

• A command-line interface.

Long term

The goal is a complete Python interface to DWD Open Data:

• Observations and current station reports (weather/weather_reports).
• Radar: composites and single-site products (weather/radar).
• Point forecasts: MOSMIX (weather/local_forecasts).
• Weather warnings (weather/alerts).
• Satellite products (weather/satellite).

License

MIT. See LICENSE.

Forecast data itself is provided by Deutscher Wetterdienst (DWD) under its own terms of use.