https://github.com/ai2cm/ace

Ai2 Climate Emulator
https://github.com/ai2cm/ace

Last synced: 3 months ago
JSON representation

Ai2 Climate Emulator

• Host: GitHub
• URL: https://github.com/ai2cm/ace
• Owner: ai2cm
• License: apache-2.0
• Created: 2023-11-29T23:08:42.000Z (11 months ago)
• Default Branch: main
• Last Pushed: 2024-08-01T21:39:55.000Z (3 months ago)
• Last Synced: 2024-08-10T01:37:53.081Z (3 months ago)
• Language: Python
• Homepage:
• Size: 3.52 MB
• Stars: 29
• Watchers: 4
• Forks: 7
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• open-sustainable-technology - ACE - A 200M-parameter, autoregressive machine learning emulator of an existing comprehensive 100-km resolution global atmospheric model. (Atmosphere / Atmospheric Composition and Dynamics)

README

        
[![Docs](https://readthedocs.org/projects/ai2-climate-emulator/badge/?version=latest)](https://ai2-climate-emulator.readthedocs.io/en/latest/)

# ACE: Ai2 Climate Emulator

This repo contains code accompanying "ACE: A fast, skillful learned global atmospheric model for climate prediction" ([arxiv:2310.02074](https://arxiv.org/abs/2310.02074)).

## Documentation

See complete documentation [here](https://ai2-climate-emulator.readthedocs.io/en/latest/).

## Quickstart

A quickstart guide may be found [here](https://ai2-climate-emulator.readthedocs.io/en/latest/quickstart.html).

## Model checkpoint

The trained ACE checkpoint and a 1-year subsample of the validation dataset is available in

[this Zenodo repository](https://doi.org/10.5281/zenodo.10791087).

## Available datasets

Two versions of the complete dataset described in [arxiv:2310.02074](https://arxiv.org/abs/2310.02074)

are available on a requester pays Google Cloud Storage bucket:

```

gs://ai2cm-public-requester-pays/2023-11-29-ai2-climate-emulator-v1/data/repeating-climSST-1deg-zarrs

gs://ai2cm-public-requester-pays/2023-11-29-ai2-climate-emulator-v1/data/repeating-climSST-1deg-netCDFs

```

The `zarr` format is convenient for ad-hoc analysis. The netCDF version contains our

train/validation split which was used for training and inference.