Saved in:
| Main Author: | |
|---|---|
| Format: | Recurso digital |
| Language: | |
| Published: |
Zenodo
2026
|
| Online Access: | https://doi.org/10.5281/zenodo.18245206 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Table of Contents:
- <h2>Introduction</h2> <p>This repository contains the figures and codes confirmed for the master thesis</p> <blockquote> <p><strong>Estimation of Geomagnetic Secular Variation by Machine Learning with Extended Kalman Filter</strong></p> <p>by Sho SATO,<br>submitted to the Graduate School of Science, Kyoto University,<br>in fulfillment of the requirements for a Master's degree,<br>April 2026.</p> </blockquote> <p>The code is designed to estimate future geomagnetic secular variation by training on time series of geomagnetic field data and Earth's Length-Of-Day data using machine learning techniques.</p> <p>For more details on the methodology, please refer to the paper (to appear in <em>Earth, Planets and Space</em>):</p> <blockquote> <p><strong>Recurrent neural network trained with the extended Kalman filter to forecast the geomagnetic secular variation for IGRF-14</strong></p> </blockquote> <h2>Files</h2> <p>This repository consists of the following four directories:</p> <ol> <li><code>code</code></li> <li><code>data</code></li> <li><code>output</code></li> <li><code>visualization</code></li> </ol> <h3>code</h3> <p>This directory contains python code files used to process data and train models using machine learning. The hidden node size of RNN is set to $D_\mathbf{h} = 34$.</p> <ul> <li> <p><code>2024_0912_processMCM2024.ipynb</code>: A script that computes differences in the provided magnetic field data and converts them to <code>.csv</code> and <code>.npy</code> formats</p> </li> <li> <p><code>2025_1017_yBnLODnLDT_h34_s0-32.py</code>: A script for performing exhaustive grid search of</p> <ul> <li>Order of derivative $d$ in the range of $0 \leq d \leq 4$</li> <li>Initial states $\mathbf{w}_0^s$ in the range of $00000 \leq s \leq 11111$,</li> </ul> <p>Training is performed using</p> <ul> <li>MCM-2024</li> </ul> </li> <li> <p><code>2025_1105_processLODdata.ipynb</code>: A script that computes moving-average filtered LOD data and variances.</p> </li> <li> <p><code>2025_1108_yByLODnLDT_h34_s0-32.py</code>: A script for performing exhaustive grid search of</p> <ul> <li>Order of derivative $d$ in the range of $0 \leq d \leq 4$</li> <li>Initial states $\mathbf{w}_0^s$ in the range of $00000 \leq s \leq 11111$,</li> </ul> <p>Training is performed using</p> <ul> <li>MCM-2024</li> <li>LOD data</li> </ul> </li> <li> <p><code>2025_1205_yBnLODyLDT_h34_s0-32.py</code>: A script for performing exhaustive grid search of</p> <ul> <li>Order of derivative $d$ in the range of $0 \leq d \leq 4$</li> <li>Initial states $\mathbf{w}_0^s$ in the range of $00000 \leq s \leq 11111$,</li> </ul> <p>Training is performed using</p> <ul> <li>MCM-2024</li> <li>First time derivative of LOD data</li> </ul> </li> </ul> <h3>data</h3> <p>This directory contains training data used for machine learning.</p> <ol> <li><strong>geomagnetic field snapshots (gauss coefficients derived from MCM-2024 model)</strong> provided by the <em>Institut de Physique du Globe de Paris</em> (IPGP) in France.</li> <li><strong>Observed Length-Of-Day (LOD) data</strong> provided by the <em>International Earth Rotation and Reference Systems Service</em> (IERS).</li> </ol> <p>The data are used as training inputs for the machine learning models.</p> <ul> <li><code>raw/</code>: Raw data as originally provided by IPGP and IERS</li> <li><code>processed/</code>: Preprocessed data, where magnetic field differences have been computed and saved in <code>.csv</code> and <code>.npy</code> formats</li> </ul> <h3>output</h3> <p>This directory contains the output results of the RNN models.</p> <h3>visualization</h3> <p>This directory contains jupyter notebooks for visualizing the results presented in the manuscript:</p> <ul> <li> <p><code>visualization/2024_0912_displayMCM2024.ipynb</code>: Notebook for visualizing MCM-2024 model data.</p> </li> <li> <p><code>visualization/2025_1105_processLODdata.ipynb</code>: Notebook for processing LOD data (used in <code>code/2025_1108_yByLODnLDT_h34_s0-32.py</code> and <code>code/2025_1205_yBnLODyLDT_h34_s0-32.py</code>).</p> </li> <li> <p><code>visualization/2025_1111_vizMCM-RNN_derivative.ipynb</code>: Notebook for visualizing the results of RNN trained with MCM-2024 model with different orders of derivatives (for Chapter 3).</p> <ul> <li>Other notebooks for visualizing the results of RNN trained with MCM-2024 model with different initial states are available on the following repository: <blockquote> <p>Sato, S., Lesur, V., Nakano, S., Minami, T., Matsushima, M., & Toh, H. (2025). IGRF-14 Japanese Candidate Model. Zenodo. https://doi.org/10.5281/zenodo.15726524</p> </blockquote> </li> </ul> </li> <li> <p><code>visualization/2025_1112_viz2015SApulse_analysis.ipynb</code>: Notebooks for visualizing the results of RNN trained with MCM-2024 model, RNN trained with 2015 SA data, and RNN trained with GAP data, respectively (for Chapter 4).</p> <ul> <li><code>visualization/2025_1114_viz2015MCM_Br.ipynb</code></li> <li><code>visualization/2025_1113_viz2015RNN_Br.ipynb</code></li> <li><code>visualization/2025_1113_viz2015GAP_Br.ipynb</code></li> </ul> </li> <li> <p><code>visualization/2025_1209_vizMCM_LOD_LDT-RNN.ipynb</code>: Notebook for visualizing the results of RNN trained with MCM-2024 + LOD data + first time derivative of LOD data (for Chapter 4 and 5).</p> </li> </ul> <h2>Closing Remarks</h2> <h3>Execution Environment</h3> <p>The code in this repository can be executed using the Python environment described in <a href="environment.yml">environment.yml</a>.</p> <p>However, the notebook <code>visualization/2025_11xx_viz2015XXX_Br.ipynb</code> requires a different virtual environment, as described in the <a href="https://github.com/IAGA-VMOD/IGRF14eval/blob/main/README.md#local-development">IAGA tutorial</a>:</p> <p><a href="https://github.com/IAGA-VMOD/IGRF14eval/blob/main/environment-base.yml">https://github.com/IAGA-VMOD/IGRF14eval/blob/main/environment-base.yml</a></p> <h3>License</h3> <p>Code in this repository is licensed under MIT, while data and documentation are licensed under CC BY 4.0. Refer to the repository LICENSE files for details.</p>