OneScience-Group/ACE2-Seasonal

ACE2-Seasonal

Model Introduction

ACE2-Seasonal reproduces the global seasonal hindcast experiment using ACE2-ERA5. Six-hour autoregression and persistent SST and sea-ice anomalies form a 64-member lagged ensemble.

Paper: Skilful global seasonal predictions from a machine learning weather model trained on reanalysis data
https://doi.org/10.1038/s41612-025-01198-3

Model Description

The study was conducted by teams at the Met Office, University of Exeter, and Ai2. ACE2 was trained with ERA5 atmospheric reanalysis, while seasonal experiments use ERA5 initial states and persistent ocean-boundary anomalies. It supports global DJF prediction and NAO, ENSO-teleconnection, and ensemble spread-skill analysis.

Use Cases

Use Case	Description
Seasonal ensembles	Construct a 64-member lagged ensemble.
NAO prediction	Evaluate ensemble-mean NAO correlation and spread.
Boundary forcing	Persist initialized SST and sea-ice anomalies.
ModelScope/OneCode execution	Validate structured data, training, inference, seasonal metrics, and visualization.
Multi-GPU training	Start multi-process training through torchrun.

Usage Instructions

hf download OneScience-Group/ACE2-Seasonal --local-dir ./ACE2-Seasonal
cd ACE2-Seasonal

Environment Dependencies

Hardware Requirements

• A GPU or DCU is recommended.
• A CPU can be used for connectivity validation with the default small-sample configuration.
• DCU users should install DTK 25.04.2 or a compatible OneScience-recommended version first.

DCU Environment

# Activate DTK and Conda first
conda create -n onescience311 python=3.11 -y
conda activate onescience311
pip install onescience[earth-dcu] -i http://mirrors.onescience.ai:3141/pypi/simple/ --trusted-host mirrors.onescience.ai

GPU Environment

# Activate Conda first
conda create -n onescience311 python=3.11 -y libstdcxx-ng=12 libgcc-ng=12 gcc_linux-64=12 gxx_linux-64=12
conda activate onescience311
pip install onescience[earth-gpu] -i http://mirrors.onescience.ai:3141/pypi/simple/ --trusted-host mirrors.onescience.ai

python scripts/fake_data.py
python scripts/train.py
torchrun --nproc_per_node=2 --nnodes=1 --master_addr="localhost" --master_port=29500 scripts/train.py
python scripts/inference.py
python scripts/result.py

Synthetic data retain the global-grid protocol, boundary variables, six-hour order, and 64-member ensemble. The six-hour state-transition target completes backpropagation and parameter updates in verified single-process and two-process DDP runs. Inference restores the checkpoint, performs 12 engineering forecast steps, and produces a finite [64,12,8,16,16] ensemble while retaining the incomplete-coverage marker. Training results are saved to result/checkpoints/ace2_seasonal.pt; ensemble predictions and seasonal metrics are saved under result/output/ and result/evaluation/.

Official OneScience Information

Platform	OneScience Main Repository	Skills Repository
Gitee	https://gitee.com/onescience-ai/onescience	https://gitee.com/onescience-ai/oneskills
GitHub	https://github.com/onescience-ai/OneScience	https://github.com/onescience-ai/oneskills

Citation and License

This repository is an independent engineering reproduction of the public ACE2-Seasonal specifications.

The original paper is licensed under CC BY 4.0; the paper, ACE2 code, weights, and ERA5 data remain subject to their respective licenses and terms.