[ICCV 25] Deep Space Weather Model: Long-Range Solar Flare Prediction from Multi-Wavelength Images 🛰️

Deep Space Weather Model (Deep SWM) - accurately and reliably predicts solar flares by capturing long-range spatio-temporal dependencies and fine-grained features in multi-channel solar image series.

📢 News

• [2026-04] 🎉 The full FlareBench dataset (~1.26 TB, covering the complete 11-year solar cycle) is now publicly available. See 📊 Data Preparation for the download URL and usage notes.

📖 Abstract

Click to expand

Accurate, reliable solar flare prediction is crucial for mitigating potential disruptions to critical infrastructure, while predicting solar flares remains a significant challenge. Existing methods based on heuristic physical features often lack representation learning from solar images. On the other hand, end-to-end learning approaches struggle to model long-range temporal dependencies in solar images.

In this study, we propose Deep Space Weather Model (Deep SWM), which is based on multiple deep state space models for handling both ten-channel solar images and long-range spatio-temporal dependencies. Deep SWM also features a sparse masked autoencoder, a novel pretraining strategy that employs a two-phase masking approach to preserve crucial regions such as sunspots while compressing spatial information. Furthermore, we built FlareBench, a new public benchmark for solar flare prediction covering a full 11-year solar activity cycle, to validate our method. Our method outperformed baseline methods and even human expert performance on standard metrics in terms of performance and reliability.

---

✨ Key Features

• Solar Spatial Encoder (SSE): Captures spatio-temporal features by selectively weighting image channels and modeling dependencies across space and time.
• Long-range Temporal SSM (LT-SSM): Extends deep state-space models to capture temporal patterns exceeding the solar rotation period.
• Sparse MAE: A tailored pretraining strategy for solar images that preserves critical but sparse regions (like sunspots) using a novel two-phase masking approach.
• FlareBench Dataset: A new public benchmark covering a complete 11-year solar cycle for robust and unbiased evaluation.

Deep SWM Architecture: Solar Spatial Encoder (SSE) and Long-range Temporal SSM (LT-SSM)

Our Sparse MAE pre-training preserves critical sunspot regions better than standard methods.

---

🏆 Results

Our model demonstrates state-of-the-art performance, outperforming existing methods and even human experts on the comprehensive FlareBench dataset.

Quantitative Comparison

Our method achieves the highest scores across all standard evaluation metrics.

Method	Test Period	GMGS↑	BSS≥M↑	TSS≥M↑
Flare Transformer (w/o PF) [1]	2014-2017 (4 years)	0.220±0.116	-1.770±0.225	0.198±0.371
DeFN-R [2]	2014-2015 (2 years)	0.302±0.055	0.036±0.982	0.279±0.162
CNN-LSTM	2019-12-01 - 2022-11-30 (3 years)	0.315±0.166	0.272±0.259	0.330±0.306
DeFN [3]	2014-2015 (2 years)	0.375±0.141	0.022±0.782	0.413±0.150
Flare Transformer (full) [1]	2014-2017 (4 years)	0.503±0.059	0.082±0.974	0.530±0.112
Ours	2019-12-01 - 2022-11-30 (3 years)	0.582±0.032	0.334±0.299	0.543±0.074
Experts	2000-2015 (16 years)	0.48	0.16	0.50

Qualitative Examples

Deep SWM successfully predicts high-impact X-class flares where baseline models fail.

Qualitative results showing successful predictions of X-class and M-class solar flares.

---

🛠️ Getting Started

Installation

Clone the repository and set up the environment:

# Clone the repository
git clone https://github.com/username/DeepSWM.git
cd DeepSWM

# Install dependencies
pip install -r requirements.txt

Usage

📊 Data Preparation

The full FlareBench dataset (approximately 1.26 TB, tar + zstd compressed) is available for academic use at the following URL:

https://pub-4f7db9ce770f41a091c81b96648d9afd.r2.dev/flarebench_dataset.tar.zst

The archive contains the raw aia/, hmi/, and xrs/ directories. After downloading, extract it with

tar --zstd -xf flarebench_dataset.tar.zst

Then build the hourly HDF5 datasets with ml/src/data/create_datasets.py. The script expects aia/ and hmi/ to be located directly under --data_root, while xrs/ is expected to reside in the parent directory of --data_root (see the path construction in the script). Please arrange the extracted directories accordingly before running the script.

The dataset is provided for academic use only. Please do not redistribute it.

🚀 Pre-training

Run the pre-training process to learn representations from the magnetogram images:

cd ml
python pretrain.py --input_dir ../flarebench_dataset/all_data_hours \
                  --output_dir ../flarebench_dataset/all_features \
                  --mode train \
                  --data_root ../flarebench_dataset

🔍 Feature Extraction

Extract intermediate features from the pre-trained model:

cd ml
python pretrain.py --input_dir ../flarebench_dataset/all_data_hours \
                  --output_dir ../flarebench_dataset/all_features \
                  --mode inference_all \
                  --data_root ../flarebench_dataset

Organize the extracted features by sample:

cd ml
python src/features/create_history_features.py

🎯 Training

Train the model using the extracted features:

cd ml
python main.py --params params/main/params.yaml \
              --imbalance \
              --fold 3 \
              --data_root ../flarebench_dataset \
              --cuda_device 0

📚 References

[1] Kanta Kaneda et al. Flare transformer: Solar flare prediction using magnetograms and sunspot physical features. In ACCV, pages 1488–1503, 2022.

[2] Naoto Nishizuka, Yuki Kubo, et al. Reliable probability forecast of solar flares: Deep flare net-reliable (defn-r). The Astrophysical Journal, 899(2):150, 2020.

[3] N. Nishizuka, K. Sugiura, Y. Kubo, M. Den, and M. Ishii. Deep flare net (defn) model for solar flare prediction. The Astrophysical Journal, 858(2):113, 2018.

📝 Citation

If you find our work helpful, please consider citing the following paper and/or ⭐ the repo:

@inproceedings{nagashima2025deepswm,
  title={Deep Space Weather Model: Long-Range Solar Flare Prediction from Multi-Wavelength Images},
  author={Shunya Nagashima and Komei Sugiura},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  year={2025}
}

📄 License

This work is licensed under the BSD-3-Clause-Clear License. To view a copy of this license, see LICENSE.