1. Quickstart Guide

This section provides a recipe for an end-to-end run of nested- and global-EAGLE on Ursa. To run a global configuration with this quickstart guide, replace references to nested with global. At present, Ursa is the only supported platform. Future development will include additional platforms.

Note

GNU make version 3.82 or higher is required.

Note

The EAGLE runtime software environment currently requires over 50 GB of disk space. Consider available space, quota, etc. when choosing where to clone the EAGLE repository and run the following steps.

If you do not already have the repository, clone it and change into the repository root:

git clone https://github.com/NOAA-EPIC/EAGLE.git
cd EAGLE

Complete the following steps from the repository root.

1.1. Building and Running EAGLE

1. Create all environments

   make env cudascript=ursa
   

   This step creates the runtime software environment, comprising conda virtual environments to support data preparation, training, inference, and verification. The conda/ subdirectory it creates is self-contained and can be removed and recreated by running the make env command again, as long as pipeline steps are not currently running.

   Developers who will be modifying Python driver code should replace make env with make devenv, which will create the same environments but also install additional code-quality tools for formatting, linting, shellchecking, typechecking, unit testing, and YAML linting.

   Note

   EAGLE virtual environments are built using both conda and pip packages. If you examine the output from the make env command above, you may see messages like the following, from pip:

   ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts.
   

   This will be followed by a report of packages pip believes are not installed or are installed with incompatible versions. These messages are due to fundamental differences in how conda and pip operate, and are generally safe to ignore. But please open an Issue if you later encounter problems you believe are related to package versions.

2. Create the EAGLE YAML config

   make config compose=base:nested:ursa >eagle.yaml
   

   The config target operates on .yaml files in the config/ directory, so this command composes config/base.yaml, config/nested.yaml, and config/ursa.yaml and redirects the composed config into eagle.yaml.

3. Set the app.base value in eagle.yaml to the absolute path to the current (repository root) directory.

   The run directories from subsequent steps, along with the output of those steps, will be created in the run/<expname> subdirectory of app.base, where <expname> is the value of app.experiment_name.

   Verify the app.account value. The default configuration sets app.account to epic. If you do not have access to the epic account on Ursa, update this value to an account you are authorized to use.

4. Create training data

   make data config=eagle.yaml
   

   This step provisions data required for training and inference. The data target delegates to targets grids-and-meshes, zarr-gfs, and zarr-hrrr, which can also be run individually (e.g. make grids-and-meshes config=eagle.yaml), but note that grids-and-meshes, which runs locally, must be run first. The zarr-gfs and zarr-hrrr targets can be run in quick succession, as they submit batch jobs: Do not proceed until their batch jobs complete successfully (see the files run/<expname>/data/*.out).

5. Train the ML model

   make training config=eagle.yaml
   

   This step trains a model using data provisioned by the previous step. It submits a batch job; do not proceed until the batch job completes successfully (see the file run/<expname>/training/runscript.training.out).

6. Run inference

   make inference config=eagle.yaml
   

   This step performs inference, producing a forecast. It submits a batch job. Do not proceed until the batch job completes successfully (see the file run/<expname>/inference/runscript.inference.out.)

7. Model verification

   make vx-grid-global config=eagle.yaml
   make vx-grid-lam config=eagle.yaml
   make vx-obs-global config=eagle.yaml
   make vx-obs-lam config=eagle.yaml
   

   For running just the global verification run, only submit vx-grid-global and vx-obs-global.

   Before running verification, the WXVX driver will run prewxvx to prepare forecast output from the previous step. See the files run/<expname>/vx/prewxvx/{global,lam}/runscript.prewxvx-*.out for details.

   These steps perform verification of the global or LAM forecasts against gridded analyses (*-grid-*) or PrepBUFR observations (*-obs-*) as truth. Each submits a batch job, so the four make commands can be run in quick succession to get all the batch jobs running in parallel. When each batch job completes, MET .stat files and .png plot files can be found under the stats/ and plots/ subdirectories of run/<expname>/vx/grid2{grid,obs}/{global,lam}/run/. The files run/<expname>/vx/*.log contain the logs from each verification run.

8. Make additional visualization outputs

   make vis-grid-global config=eagle.yaml
   make vis-grid-lam config=eagle.yaml
   make vis-obs-global config=eagle.yaml
   make vis-obs-lam config=eagle.yaml
   

   For running just the global visualization run, only submit vis-grid-global and vis-obs-global.

   These steps will first call eagle-tools’s postwxvx tool to create and save a series of netCDF files with all relevant statistics in the corresponding wxvx directory for each variable. It will then create a series of basic plots (provided by DataArray.plot() from the xarray library) in the run/<expname>/visualization/grid2{grid,obs}/{global,lam}/plots-basic directory.

   For the grid-based vis-grid-global and vis-grid-lam targets, additional error plots (forecast vs truth differences) will be created under run/<expname>/visualization/grid2grid/{global,lam}/plots-spatial-stats/. These plots depend on 1. The config value at key-path vx.grid2grid.{global,lam}.wxvx.wxvx.ncdiffs being set to true, which instructs MET to produce netCDF difference files during verification; and 2. The config block at key-path visualization.grid2grid.{global,lam}.visualization.spatial_stat_plots, which enables and configures plot generation, being present.

9. Run inference in near-real-time (NRT)

   1. Create the EAGLE NRT config

   make config compose=base:nested:ursa:nrt-nested > nrt-composed.yaml
   

   2. Set the app.base value in nrt-composed.yaml to the absolute path to the current (repository root) directory.

   This should match the path used when generating the main EAGLE config above.

   Two additional paths may require attention:

   • inference.anemoi.checkpoint_dir

   • grids_and_meshes.rundir

   If you are following only the quickstart workflow, you do not need to modify these values. The config automatically pulls both paths from the quickstart run. However, if you ran multiple experiments or stored outputs in a different location, update these paths so they point to the correct directories.

   3. Realize the EAGLE NRT config

   make realize config=nrt-composed.yaml > nrt.yaml
   

   This creates the final config to begin a NRT run. It is required because it freezes the NOW environment variable across the entire configuration. Since jobs may be submitted at different times, this ensures a consistent timestamp is used throughout the run.

   4. Load current initial conditions

   make data config=nrt.yaml
   

   5. Run inference

   make inference config=nrt.yaml
   

   Your forecast will save to path/to/eagle/src/run/default/nrt_inference/YYYY/MM/DD/HH/inference.

   6. Verify your forecast

   make vx-grid-global config=nrt.yaml
   make vx-grid-lam config=nrt.yaml
   make vx-obs-global config=nrt.yaml
   make vx-obs-lam config=nrt.yaml
   

   Verification requires completed forecast output, so wait until the inference window has completed before running these commands. For example, if you created a 48 hour forecast, you will need to wait 48 hours to verify it. Once the output is available, run the same vx targets shown above to verify against gridded analyses or observations.