MP-DualNet: A Dual-Branch Deep Learning Approach for Magnetopause Boundary Detection Using Simulated SXI Images

Overview

This repository provides the dataset, trained model, and source code for MP-DualNet, a dual-branch deep learning framework for automated magnetopause detection from simulated soft X-ray imager (SXI) observations.

The repository supports full reproducibility of the results presented in the associated paper.

Repository Structure

MP-DualNet/
├── Data/
│ ├── train_val_test.npz
│ ├── storm_2024_event.npz
│
├── Model/
│ ├── best_DualNet.weights.h5
│
├── Code/
│ ├── train.py
│
├── Notebook/
│ ├── visualization.ipynb

Dataset Description

1. Training Dataset (train_val_test.npz)

This file contains the dataset used for training, validation, and testing.

Contents:

X: SXI images (input)

Y: Magnetopause envelope masks (ground truth)

Data format:

X: shape (N, H, W, 1)

Y: shape (N, H, W, 1)

The dataset includes predefined splits for:

training

validation

testing

2. Storm Event Dataset (storm_2024_event.npz)

This dataset is used for independent evaluation.

Contents:

Simulated SXI images under 2024 geomagnetic storm conditions

Corresponding magnetopause envelope masks

Purpose:

Evaluate model generalization under disturbed space weather conditions

Model

The trained model is provided at:

Model/best_DualNet.weights.h5

This corresponds to the best-performing checkpoint reported in the paper.

How to Reproduce Results

1. Training

Run:

python Code/train.py

2. Inference and Visualization

Open the notebook:

Notebook/visualization.ipynb

Run all cells to:

load the trained model

perform inference

generate visualization results

Requirements

Python 3.11

Tensorflow

NumPy

Matplotlib

(Optional but recommended)

Jupyter Notebook

Method Reference

The model implementation builds upon the framework described in:

Liu et al. (2025), A&A, 698, A263https://doi.org/10.1051/0004-6361/202453627

with task-specific modifications for magnetopause detection.

Data Generation

The SXI dataset is generated from:

global hybrid simulations

SXI forward modeling

Zhongwei Yang and Tianran Sun conduct the simulations.

Reproducibility

All components required to reproduce the results are included:

dataset

trained model

source code

No additional proprietary data is required.

Notes

The model is designed to capture large-scale magnetopause morphology.

Small-scale boundary variations may not be fully resolved due to resolution and noise limitations.

Contact

For questions or collaboration, please contact the authors(Email: [e-mail], [e-mail]).