Towards calibrated ensembles of neural weather model forecasts

This repository contains the material and guidelines to reproduce the results presented in the manuscript entitled Towards calibrated ensembles of neural weather models, submitted to Journal of Advances in Modeling Earth Systems. We provide scripts that illustrate how to perform inference with bred vectors for a single AFNO model. Instructions to train AFNO models are available from Pathak et al., 2022. The repository is structured in two folders:

• scripts --> Main executable scripts.
  • scripts-inference --> Python scripts to perform inference with a neural weather model by perturbing the initial condition with white noise (inference_G-AFNO.py) or bred noise (inference_EnAFNO.py). We also provide a python script to compute bred vectors for neural weather models (compute_bred-vectors.py).
  • scripts-download-era5 --> Python scripts to download ERA5 from the Copernicus climate data store. Note that you have to register for an account at the European Center for Medium-Range Weather Forecasts (ECMWF) to download the data.
• utils --> This folder contains the auxiliary scripts that are sourced by the main scripts during execution.