Wei Wang1,WeiDong Yang1,Lei Wang2, GuiHua Wang2, Lei RuiBo3,
1 School of Computer Science, Fudan University, Shanghai, China, 2 Department of Atmospheric and Oceanic Sciences & Institute of Atmospheric Sciences, Fudan University, Shanghai, China, 3 Key Laboratory of Polar Science, MNR, Polar Research Institute of China, Shanghai, China.
Paper: https://doi.org/10.1038/s41612-025-01058-0
Paper downloading
Supplementary Information: https://doi.org/10.1038/s41612-025-01058-0
Supplementary Information downloading
The rapid decline of Arctic sea ice resulting from anthropogenic climate change poses significant risks to indigenous communities, ecosystems, and the global climate system. This situation emphasizes the immediate necessity for precise seasonal sea ice forecasts. While dynamical models perform well for short-term forecasts, they encounter limitations in long-term forecasts and are computationally intensive. Deep learning models, while more computationally efficient, often have difficulty managing seasonal variations and uncertainties when dealing with complex sea ice dynamics. In this research, we introduce IceMamba, a deep learning architecture that integrates sophisticated attention mechanisms within the state space model. Through comparative analysis of 25 renowned forecast models, including dynamical, statistical, and deep learning approaches, our experimental results indicate that IceMamba delivers excellent seasonal forecasting capabilities for Pan-Arctic sea ice concentration. Specifically, IceMamba outperforms all tested models regarding average RMSE and anomaly correlation coefficient (ACC) and ranks second in Integrated Ice Edge Error (IIEE). This innovative approach enhances our ability to foresee and alleviate the effects of sea ice variability, offering essential insights for strategies aimed at climate adaptation.
IceMamba: Arctic Sea Ice Forecasting Framework
Developed through multi-institutional collaboration led by the School of Computer Science at Fudan University, in partnership with the Department of Atmospheric Sciences and China Polar Research Institute, IceMamba is an open-source deep learning framework specializing in pan-Arctic seasonal sea ice concentration (SIC) forecasting.
We provide full accessibility to accelerate polar climate research:
- Model Weights: Pretrained parameters for rapid deployment
- Training Datasets: Processed SIC observations with spatiotemporal metadata
- Configuration Files: Pre-optimized experimental setups
🔗 Persistent Access:
All assets are permanently hosted on Zenodo with version control：https://zenodo.org/records/14926245
📜 Licensing:
- Data: CC-BY 4.0 International
- Code: MIT License
git https://github.com/WeiWang31/IceMamba.git
cd IceMambaThe base environment is cuda 11.7=11.7, python=3.7
- create conda environment
conda create -n icemamba python=3.7
conda activate icemamba- Install Dependencies
pip install -r requirements.txt
pip install torch==1.13.0+cu117 torchvision==0.14.0+cu117 torchaudio==0.13.0 --extra-index-url https://download.pytorch.org/whl/cu117
pip install pytorch-lightning==1.9.5
pip install torchsummary
cd selective_scan && pip install . && pytest🔗 Download the complete resource bundle from our Zenodo repository:
Package includes:
- Raw observational datasets
- Pretrained model checkpoints (CKPT)
# Unpack archives and configure paths
unzip ERA5_EASE.zip
unzip ORAS5_EASE.zip
unzip ckpt.zip
unzip IceMamba-4_SIPN_ckpt.zip
unzip IceMamba-1-only-SIC_SIPN_ckpt.zip
# Data deployment
mv ERA5_EASE ORAS5_EASE ./data_preprocess/
# Model deployment
mv ckpt/ ./Model/IceMamba/
mv IceMamba-4_SIPN_ckpt/ ./SIPN_evaluation_IceMamba-4
mv IceMamba-1-only-SIC_SIPN_ckpt/ ./SIPN_evaluation_IceMamba-1-only-SICcd ./data_preprocess
python data_preprocess.py- Testing all IceMamba variants
cd ./Model/IceMamba
bash test.sh- Testing the IceMamba-4 for SIPN benchmark
cd ./SIPN_evaluation_IceMamba-4
bash SIPN_evaluation.sh
python calculate_SIPN_evaluation_result.py- Testing the IceMamba-1-only-SIC for SIPN benchmark
cd ./SIPN_evaluation_IceMamba-1-only-SIC
python IceMamba-1-only_SIC_test.py
python calculate_SIPN_evaluation_result.pycd ./explainability
# generate two test sets
python generate_testset.py
# Explainability Test
python permute_test.py
# calculate MAE after permuting
python permute_result_analysis_month.py
python permute_result_analysis_month_ua10.pyIf you want to train IceMamba variants like in our paper. You can:
cd ./Model/IceMamba
bash train_iceMamba-1.sh
bash train_iceMamba-4.sh
bash train_iceMamba-6.shIf you want to train IceMamba-4 for SIPN benchmark like in our paper. You can:
cd ./SIPN_evaluation_IceMamba-4
bash train.shIf you want to train IceMamba-1-only-SIC for SIPN benchmark like in our paper. You can:
cd ./SIPN_evaluation_IceMamba-1-only-SIC
bash train.shLeverage our framework to develop novel IceMamba variants or entirely new sea ice prediction architectures. Implementation workflow:
🛠 Configuration Guide
- Preprocessing Control (preprocess_dict_xx.json)
- "abs": Standard normalization
- "anomaly": Anomaly-normalized hybrid processing
- Critical: Variables must maintain strict alignment with time_dict_xx.json
- Temporal Configuration (
time_dict_xx.json)- Defines historical windowing strategy through lag_month parameters
- Supports dynamic temporal dependency engineering
Overall, IceMamba is not a single-purpose model, but a modular deep learning framework specifically designed for flexible sea ice prediction across multiple temporal scales (e.g., short-term, seasonal, and long-term forecasting). Key features include:
- 🧩 Decoupled Architecture: Our implementation with PyTorch Lightning creates clean separation between model components and training workflow
- 🔧 Customizable Framework: Researchers can either:
a) Modify/replace the core IceMamba model
b) Directly adopt our training infrastructure for custom sea ice prediction models - 🤝 Research Community: We actively encourage academic collaborations. Feel free to:
- Build upon our codebase
- Contact us at wang_wei23@m.fudan.edu.cn for technical discussions
If this work contributes to your research, please consider leaving a ⭐️ star and citing our paper to support open scientific development.
@article{wang2025seasonal,
title={Seasonal forecasting of Pan-Arctic sea ice with state space model},
author={Wang, Wei and Yang, Weidong and Wang, Lei and Wang, Guihua and Lei, Ruibo},
journal={npj Climate and Atmospheric Science},
volume={8},
number={1},
pages={1--17},
year={2025},
publisher={Nature Publishing Group}
}This project is based on Mamba (paper), VMamba (paper, code), Swin-Transformer (paper, code)