Skip to content

Postprocessing Output Structure

When you run the canari_ml postprocess command, the output can be found under the outputs/ directory by default. The final postprocessing results include netCDF files containing the processed predictions, ready for analysis or visualisation. In the future, this postprocessing step can include other options than just generating a netCDF.

Overview

The postprocessing pipeline performs the following key steps:

  • Takes the raw model predictions generated by the canari_ml predict command.
  • Processes (convert change of state of ua700 to actual normalised ua700 for each date) then denormalises the raw predictions.
  • Saves the processed predictions into .nc (netCDF), for easy access and analysis.
  • Organises the processed outputs into a dedicated folder structure under results/.

Simplified Output Tree

The main output directory is the same as the prediction step.

outputs/
└── demo_train/                     # Parent directory corresponding to the training run
    └── prediction/
        └── 1979-01-26/             # Prediction name
            └── 42/
                ├── predict_2025-09-16_11-52-38.log  # Postprocess log
                └── results/
                    └── postprocess/
                        └── netcdf/
                            └── prediction_file.nc  # Processed predictions in netCDF format

Where:

  • postprocess/netcdf/ contains the final processed predictions in netCDF format.

Output Breakdown

prediction/<predict.name>/results/postprocess/netcdf/

  • The postprocessed prediction results are stored in this directory. The netcdf/ subfolder contains the processed predictions in netCDF format.
  • prediction_file.nc is the default netCDF output filename.
  • If an ensemble run, this will contain the ensemble mean and standard deviation.
  • The data in this file corresponds to the model’s forecast for the specified date(s) (only 1979-01-26 in this example).

Summary

After running the canari_ml postprocess command:

  • The postprocessed prediction results are placed under outputs/<train.name>/prediction/<predict.name>/results/postprocess/netcdf/.
  • netCDF files are generated, containing processed model outputs, ready for analysis.