Other data sources

In the Quickstart guide, we showed how to use the test dataset for initial conditions. However, in practice, you will likely want to use live data to initialise the model.

Note

Further details of the inputs available can be found in the inputs module and input config documentation.

Local files

While using the training dataset is a good way to get started, if running in real time you will likely want more up to date data. It is valid to regenerate the dataset with more recent data, and use that as input. To do so, use the dataset input and provide the opening configuration:

input:
  dataset:
    ... # Dataset config here

Additionally, you can use local files directly as input with grib being supported at the moment.

input:
  grib: PATH/TO/GRIB

Remote Stores

It is also possible to use remote data stores as input sources. Currently, anemoi-inference supports the Copernicus Data Store (CDS), MARS and FDB as remote data sources.

Warning

To download data from the Copernicus Data Store (cds), you need to have the cdsapi package installed and configured in your environment. See here for more information.

input:
  cds:
    dataset: ???

Note

Usage of the ECMWF opendata service is also available, accessable through a plugin,

pip install anemoi-plugins-ecmwf-inference[opendata]

Once the plugin is installed, you can use the opendata input source.

See also