A C E clim S S T

Introduction

ACE-CLIMSST is part of the Ai2 Climate Emulator (ACE) model family, which simulates atmospheric variability over various time scales, from days to centuries. It is a research tool and should not be used for operational climate predictions. The model is quick, stable, and provides a highly accurate time-mean climate compared to its target dataset.

Architecture

The ACE-climSST model is based on the output from the FV3GFS atmospheric model, forced with annually-repeating climatological sea surface temperature. It is built to offer fast inference, making it considerably faster than typical physics-based atmospheric models.

Training

ACE-climSST was trained on output data from the FV3GFS atmospheric model. It focuses on simulating atmospheric conditions while addressing long-term stability and accuracy. However, it has limitations in responding to El Niño-like variability and long-term warming trends, and it experiences minor drifts in the total dry air mass of the atmosphere.

Guide: Running Locally

To run ACE-climSST locally, follow these basic steps:

1. Clone the repository from the GitHub.
2. Ensure you have the necessary dependencies installed as per the documentation.
3. Download the model checkpoint from Zenodo or use the provided checkpoint on Hugging Face.
4. Run the model inference script to simulate atmospheric conditions.

For better performance, consider using cloud GPUs from providers like AWS, Google Cloud, or Azure.

License

The model is licensed under the Apache-2.0 license, allowing for wide use and distribution while maintaining the necessary attributions and notices.