t.in.ecmwf downloads real-time forecasts from the ECMWF open data service and imports each requested parameter as a space-time raster dataset (STRDS). It covers the IFS forecasts at 0.25 degree and, since 20 October 2026, at 0.1 degree, and the AIFS (machine learning) forecasts at 0.25 degree.
Only the requested fields are transferred. Each forecast file comes with a JSON index giving the byte offset and length of every GRIB message, and the module fetches just those messages with HTTP range requests. They are cropped to the area (by default the current region, plus one grid cell on each side), imported with r.import, and registered with their valid time.
The datasets option takes one or more of:
| Dataset | Model | Stream/type | Content | 0.1 degree |
|---|---|---|---|---|
| ifs_oper | IFS | oper/fc | Control forecast (formerly HRES) | yes |
| ifs_enfo | IFS | enfo/ef | Ensemble, 50 perturbed members | yes |
| ifs_enfo_ep | IFS | enfo/ep | Ensemble probabilities | yes |
| ifs_wave | IFS | wave/fc | Wave control forecast | yes |
| ifs_waef | IFS | waef/ef | Wave ensemble, 50 members | yes |
| ifs_waef_ep | IFS | waef/ep | Wave ensemble probabilities | yes |
| aifs_single | AIFS Single | oper/fc | ML deterministic forecast | no |
| aifs_ens | AIFS ENS | enfo/pf | ML ensemble perturbed members | no |
| aifs_ens_cf | AIFS ENS | enfo/cf | ML ensemble control member | no |
With several datasets, they are queued: each dataset is downloaded and imported completely before the next one starts. A parameter missing from one dataset (for example swh in ifs_oper, or tp in ifs_wave) is skipped there with a warning. The module fails when a dataset offers none of the parameters, or when a parameter is in none of the datasets.
resolution=0.25 (the default) uses the 0.25 degree grid, available for every dataset. resolution=0.1 uses the 0.1 degree grid (IFS only). The module fails for an AIFS dataset at that resolution. resolution=best takes 0.1 degree when the run is already published at 0.1 degree, and 0.25 degree otherwise. ECMWF publishes the 0.1 degree files 2 hours after the 0.25 degree ones, so right after a new run the 0.25 degree grid of the newest run is preferred to the 0.1 degree grid of the previous run. AIFS datasets always use 0.25 degree under best.
At 0.1 degree, ECMWF splits each step into a surface file
(-sfc) and a pressure-level file (-pl). The
module reads the surface index first and the pressure-level index only
for parameters not found there.
By default the latest complete run is used. A run counts as complete when its last step is on the server. date and time select a given run, and both must be given. The ECMWF server keeps about four days of runs. The mirrors (source=aws or source=google) keep more.
steps selects forecast steps in hours, as numbers
and ranges (0-72,96). Steps that a run does not publish are
ignored. The IFS publishes 0-144 h every 3 hours and 150-360 h every 6
hours for the 00 and 12 UTC runs, and 0-144 h every 3 hours for the 06
and 18 UTC runs. AIFS publishes 0-360 h every 6 hours. Probabilities
(_ep datasets) come as intervals (such as 0-24 h) in two files
(240 h and 360 h), and steps selects the intervals by
their end.
The -b flag (bundle) downloads every available step of the run, up to 15 days (360 h), in place of steps. Without date and time, it takes the latest run that reaches 15 days (00 or 12 UTC for the IFS, as the 06 and 18 UTC runs stop at 144 h). Combined with several datasets, the bundles are downloaded one after the other.
parameters takes ECMWF short names (2t, tp, 10u, 10v, msl, ro, swh, ...). The -l flag prints, for each dataset, the chosen resolution and run, the published steps, and every parameter with its level type, levels and size per step, then exits without downloading.
For multi-level parameters (pressure levels, soil layers),
levels restricts the levels; by default all levels of
the parameter are kept. For ensembles, members
restricts the members (1-10,25); by default all members are
kept. An ensemble at 0.1 degree is large (about 1.7 MB per member,
parameter and step before cropping), and the module reports the download
size before starting.
One STRDS is created per parameter and level, named
<output_prefix>_<dataset>_<parameter>[_<level>],
for example ecmwf_ifs_oper_2t or ecmwf_ifs_oper_t_850.
Maps are named
<strds>_<run YYYYMMDDHH>_<step> for
points in time and
<strds>_<run>_<start>_<end> for
intervals. Ensemble members go into the same STRDS and carry the
semantic label mNN, so that
ecmwf_ifs_enfo_2t.m07 selects member 7 in the temporal
modules.
Values are in the native ECMWF units (K, m, Pa, J m-2, ...), and the STRDS description gives the unit. Precipitation and snowfall are in m of water in the IFS datasets, but in kg m-2 (mm) in the AIFS datasets. The time stamps follow the meaning of each field:
The STRDS semantic type is sum for accumulated fields, max or min for extremes, and mean otherwise. An existing STRDS is replaced, with its maps, only with --overwrite.
Downloaded GRIB messages are stored under cache_dir, one file per message, by resolution, run, dataset and step. A later call with the same cache_dir reuses them, for example to import another area or to add de-accumulation. Without cache_dir, a temporary directory is used and removed at the end.
The module needs the GDAL Python bindings with the GRIB driver, and no other extra Python package. Downloads run in parallel within a dataset (nprocs, default 4). Please keep this number small: the ECMWF server is shared.
In a project that is not in WGS84, r.import reprojects the fields with the resample method, at a resolution it estimates from the grid (about 25 km or 10 km), not at the resolution of the region.
An area crossing the antimeridian is not supported. Import the east and west parts separately.
ECMWF open data is distributed under the CC-BY-4.0 licence. Credit ECMWF in products derived from it.
List what the 0.1 degree control forecast offers:
t.in.ecmwf -l datasets=ifs_oper resolution=0.1Download the 15-day forecast of temperature, precipitation and runoff for Nepal and Bangladesh at the best resolution, with precipitation and runoff per step, for the control forecast and then the AIFS:
g.region n=31 s=20 w=80 e=93 res=0.1
t.in.ecmwf -b -d datasets=ifs_oper,aifs_single resolution=best \
parameters=2t,tp,ro cache_dir=$HOME/RSDATA/ecmwf_cache
t.rast.list ecmwf_ifs_oper_tp columns=name,start_time,end_timeEnsemble members 1 to 10 of the 2 m temperature for the next 3 days, and the probability of more than 10 mm and 25 mm of rain in 24 hours:
t.in.ecmwf datasets=ifs_enfo,ifs_enfo_ep resolution=0.1 \
parameters=2t,tpg10,tpg25 members=1-10 steps=0-72
t.rast.list ecmwf_ifs_enfo_2t.m03Temperature at 850 and 500 hPa, at 0.1 degree:
t.in.ecmwf datasets=ifs_oper resolution=0.1 parameters=t levels=850,500r.import, t.create, t.in.era5, t.rast.list, t.register
ECMWF open data: real-time forecasts from IFS and AIFS, What's new at 0.1°
Yann Chemin
Available at: t.in.ecmwf source code (history)
Accessed: Sunday Nov 15 11:08:19 2026
Main index | Temporal index | Topics index | Keywords index | Graphical index | Full index
© 2003-2026 GRASS Development Team, GRASS 8.5.1dev Reference Manual