Tools and docs for
- Adapting the Bris model by moving the domain in a Bris model to another geographical area.
- run inference on the resulting model.
- utilizing the result of the inference.
- various support tools.
For information on how to use the original Bris model, you should rather look at the bris-inference docs.
This contains the neccessary components to run the bris model in anemoi inference and Forecast-in-a-Box. It consists of several parts:
- Plugins for anemoi-inference
- A tool to adapt a checkpoint so it can run in anemoi-inference
- Later, docs for how to add this to Forecast-in-a-Box will be added.
To create a forecast you need to run inference on the model. This can be done directly with anemoi-inference or by Forecast-in-a-Box (which also uses anemoi-inference)
In order to run inference, you need access to data. There are several ways to get this, but we have tested against two data services from ecmwf: Mars and polytope. If you move the domain for a bris checkpoint, as described below, you can autmatically have polytope configured as a data source for that domain. Note, however, that neither of these data sources are freely available to the public. This means that your organization will need to somehow have arranged access to these data sources for you, unless you configure other data sources.
Install anemoi-inference, e.g in an empty directory with:
uv init
uv add anemoi-inference
Create config that fits your needs. See example-config. Also, see bris-adapt for more information on configuration.
uv run anemoi-inference run <config.yaml>
The output from anemoi inference can be hard to visualize. To aid in this, we provide a tool, make-grid, to convert to a more standardized output format. It can be run like this:
uv run bris-adapt process make-grid anemoi-output.nc grid.nc
This should create a file, grid.nc
, which can be displayed in eg. diana.
TODO