Variational Mode Decomposition and Deep Learning for Geomagnetic  K-Index Prediction

Asmadi Djasman; Ahmad Musyafa; Kahfi Heryandi  Suradiradja

doi:10.30871/jaic.v10i4.13616

Authors

Asmadi Djasman Program Studi Teknik Informatika, Pascasarjana, Universitas Pamulang
Ahmad Musyafa Program Studi Teknik Informatika, Pascasarjana, Universitas Pamulang
Kahfi Heryandi Suradiradja Program Studi Teknik Informatika, Pascasarjana, Universitas Pamulang

DOI:

https://doi.org/10.30871/jaic.v10i4.13616

Keywords:

Convolutional Neural Network, Geomagnetic Storm, K-Index Prediction, Long Short-Term Memory, Variational Mode Decomposition

Abstract

Geomagnetic K-index characterizes local disturbance intensity on a quasi-logarithmic scale and serves as a critical input for space weather early warning systems. Predicting this index is challenging due to its non-stationary, chaotic nature and severely imbalanced class distribution. This study develops and compares two univariate hybrid deep learning models, VMD-CNN 1D and VMD-LSTM, for K-index prediction using data from BMKG Tuntungan Observatory from January 2020 to June 2025 (16,064 samples, 3-hour resolution). The preprocessing pipeline applies Variational Mode Decomposition (VMD) with six intrinsic mode functions, followed by an eight-timestep sliding window. An ablation study confirms that VMD substantially contributes to predictive performance, reducing RMSE by 55.4% and 52.4% for CNN 1D and LSTM respectively compared to their non-VMD counterparts with comparable parameter capacity. Both models were evaluated using RMSE, MAE, and R², with statistical significance confirmed via Wilcoxon signed-rank tests. VMD-CNN 1D achieved superior overall performance (RMSE 0.3608, R² 0.8846) compared to VMD-LSTM (RMSE 0.3889, R² 0.8660), and both substantially outperformed a Persistence baseline (RMSE 0.9643). However, under storm conditions (K ≥ 5, p = 0.024), VMD-LSTM outperformed VMD-CNN 1D (R² 0.6632 versus 0.5430) and achieved higher storm-detection recall (0.891 versus 0.812). These findings indicate that architecture choice should reflect the target operational context, with VMD-CNN 1D suited for routine monitoring and VMD-LSTM for storm-period forecasting.

Downloads

Download data is not yet available.

References

[1] P. Thaker, A. Desai, P. Panchal, D. Sutaria, M. Shah, and M. Prajapati, “Evaluating AI approaches for space weather prediction_ Strengthening satellite resilience and technology systems,” Sp. Habitat., vol. 1, no. 3, p. 100032, 2025, doi: 10.1016/j.spaceh.2025.100032.

[2] J. Matzka, C. Stolle, Y. Yamazaki, O. Bronkalla, and A. Morschhauser, “The Geomagnetic Kp Index and Derived Indices of Geomagnetic Activity,” Sp. Weather, vol. 19, no. 5, pp. 1–21, 2021, doi: 10.1029/2020SW002641.

[3] Y. Y. Shprits, R. Vasile, and I. S. Zhelavskaya, “Nowcasting and Predicting the Kp Index Using Historical Values and Real-Time Observations,” Sp. Weather, vol. 17, no. 8, pp. 1219–1229, 2019, doi: 10.1029/2018SW002141.

[4] A. R. Saputra, A. Saikhu, and B. Amaliah, “Forecasting the Geomagnetic K Index with LSTM Networks: Insights from Indonesian Data,” in 2025 15th International Conference on Information & Communication Technology and System (ICTS), Denpasar, Indonesia: IEEE, Nov. 2025, pp. 1–6. doi: 10.1109/ICTS67612.2025.11369590.

[5] W. Yu, H. Li, M. Liu, and L. Song, “A geomagnetic variation prediction method based on chaotic VMD-LSTM neural network,” Acta Seismol. Sin., vol. 46, no. 1, pp. 92–105, 2024, doi: 10.11939/jass.20220132.

[6] K. Dragomiretskiy and D. Zosso, “Variational Mode Decomposition,” IEEE Trans. Signal Process., vol. 62, no. 3, pp. 531–544, Feb. 2014, doi: 10.1109/TSP.2013.2288675.

[7] Y. Xie et al., “Enhancing GNSS Deformation Monitoring Forecasting with a Combined VMD-CNN-LSTM Deep Learning Model,” Remote Sens., vol. 16, no. 10, pp. 1–21, 2024, doi: 10.3390/rs16101767.

[8] S. Guessoum et al., “The Short-Term Prediction of Length of Day Using 1D Convolutional Neural Networks (1D CNN),” Sensors, vol. 22, no. 23, pp. 1–17, 2022, doi: 10.3390/s22239517.

[9] H. Han et al., “Hybrid Climate Forecasting: Variational Mode Decomposition and Convolutional Neural Network with Long-Term Short Memory,” Polish J. Environ. Stud., vol. 33, no. 2, pp. 1121–1134, 2024, doi: 10.15244/pjoes/172756.

[10] G. Bernoux, A. Brunet, É. Buchlin, M. Janvier, and A. Sicard, “Forecasting the Geomagnetic Activity Several Days in Advance Using Neural Networks Driven by Solar EUV Imaging,” J. Geophys. Res. Sp. Phys., vol. 127, no. 10, pp. 1–22, 2022, doi: 10.1029/2022JA030868.

[11] G. Kervalishvili, I. Michaelis, M. Korte, J. Rauberg, and J. Matzka, “A Novel Model for Forecasting Geomagnetic Indices Using Machine Learning,” Geophys. Res. Lett., vol. 52, no. 8, 2025, doi: 10.1029/2025GL114848.

[12] D. H. Kim, D. J. Kim, and S. Y. Choi, “A Variational-Mode-Decomposition-Cascaded Long Short-Term Memory with Attention Model for VIX Prediction,” Appl. Sci., vol. 15, no. 10, pp. 1–28, 2025, doi: 10.3390/app15105630.

[13] H. Sun, Z. Yu, and B. Zhang, “Research on short-term power load forecasting based on VMD and GRU,” PLoS One, vol. 19, no. 7 JULY, pp. 1–21, 2024, doi: 10.1371/journal.pone.0306566.

[14] X. Yang, J. Li, and X. Jiang, “Research on information leakage in time series prediction based on empirical mode decomposition,” Sci. Rep., vol. 14, no. 1, pp. 1–18, 2024, doi: 10.1038/s41598-024-80018-9.

[15] L. N. Boucetta, Y. Amrane, A. Chouder, S. Arezki, and S. Kichou, “Enhanced Forecasting Accuracy of a Grid-Connected Photovoltaic Power Plant: A Novel Approach Using Hybrid Variational Mode Decomposition and a CNN-LSTM Model,” Energies, vol. 17, no. 7, 2024, doi: 10.3390/en17071781.

[16] R. Feng et al., “A Hybrid VMD-SSA-LSTM Framework for Short-Term Wind Speed Prediction Based on Wind Farm Measurement Data,” Energies, vol. 19, no. 3, pp. 1–20, 2026, doi: 10.3390/en19030520.