global-EAGLE

EAGLE currently includes a prototype EAGLE model trained with global GFS data.

EAGLE configurations were provided by Tim Smith at NOAA Physical Sciences Laboratory.

Training Data

The EAGLE training dataset combines regridded global and regional forecast data.

At a glance:

  • GFS is conservatively regridded to 1 degree.

  • The training period spans 2015-02-01T06 through 2023-01-31T18.

  • The validation period spans 2023-02-01T06 through 2024-01-31T18.

  • The testing period spans 2024-02-01T06 through 2025-01-31T18.

Table 2 EAGLE input variables by category

Category

Fields

Prognostic

gh, u, v, w, t, q, sp, u10, v10, t2m, t_surface, sh2

Diagnostic

u80, v80, accum_tp using fhr=6

Forcing

lsm, orog, cos_latitude, sin_latitude, cos_longitude, sin_longitude, cos_julian_day, sin_julian_day, cos_local_time, sin_local_time, insolation

The vertical levels used in the dataset are 100, 150, 200, 250, 300, 400, 500, 600, 700, 850, 925, and 1000.

Model Architecture

The EAGLE model uses the following architecture:

  • Encoder and Decoder: Graph Transformer

  • Processor: Sliding Window Transformer

  • Latent space is a 4x coarsened data space

The graph configuration connects targets to nodes through nearest neighbors in the encoder and decoder, with encoder_knn=12 and decoder_knn=3.

The latent mesh is four times coarser than the native data resolution.

Near-Real-Time Forecasting

The global-EAGLE model can be run in near real time (NRT) using the feature/global_eagle branch in this repository. That branch includes the required dependencies (including compatible anemoi versions) and is the recommended starting point for NRT runs of global-EAGLE.

To run NRT:

  1. Check out the feature/global_eagle branch.

    git checkout feature/global_eagle

  2. EPIC hosts the checkpoint on Azure. To download the checkpoint to your machine, simply run:

    wget -O inference-last.ckpt https://eaglecheckpoints.blob.core.windows.net/eagle-checkpoints/global-eagle/era5_gdas_global_check.ckpt

  3. Follow the NRT workflow, but before running its make realize step, update:

    • app.base to the absolute path of your local repository root

    • inference.anemoi.checkpoint_dir to the checkpoint you downloaded from Azure (inference-last.ckpt)

After those updates, realize the config and continue with the remaining quickstart NRT steps for the global configuration.