Aug 2026· Frontiers in Astronomy and Space Sciences· 0 citations· 21 references
TL;DR
It is shown that high-frequency fluctuations at a single station are dominated by instrumental and stochastic noise, making temporal filtering via downsampling essential, and utilizing a learned adjacency matrix as an adaptable structural regularization provides a highly stable and effective framework for short-term geomagnetic field forecasting.
Abstract
Geomagnetic field forecasting is critical for mitigating space weather hazards, yet single-station prediction remains a challenge due to the complex, non-linear coupling of vector components. In this work, we propose a Graph Neural Network (GNN) architecture enhanced with Temporal Convolutional Networks (TCN) to forecast the H, D, and Z components. By modeling the observatory’s sensors as nodes in a learned directed graph, the system captures dynamic spatio-temporal correlations between orthogonal components. We introduce a dilated inception layer to efficiently capture multi-scale temporal patterns. Experimental results using 1-min downsampled data from the MAGDAS station (Ecuador) show that our model achieves a Mean Absolute Error (MAE) of 0.8036 nT, outperforming Vector Autoregression (VAR) and thoroughly optimized LSTM baselines. While the Naive Persistence baseline yields a marginally lower global MAE due to the statistical dominance of quiet-time periods, the proposed GNN-TCN provides vastly superior phase tracking
(
R
=
0.9950
)
and structural stability. Rigorous evaluation under varying space weather conditions reveals robust performance, maintaining a highly controlled error even during the top 5% most severe local geomagnetic storms. These findings, supported by power spectral density analysis, indicate that high-frequency fluctuations at a single station are dominated by instrumental and stochastic noise, making temporal filtering via downsampling essential. Ultimately, this proof-of-concept study demonstrates that utilizing a learned adjacency matrix as an adaptable structural regularization, rather than extracting a fixed physical law, provides a highly stable and effective framework for short-term geomagnetic field forecasting.
This paper proposes FAST-GNN, a graph neural network that integrates frequency-aware and spatio-temporal adaptive mechanisms to address the challenges of modeling complex spatio-temporal dependencies in multivariate time series forecasting. Traditional GNNs often struggle with dynamic graph construction, frequency-domain feature extraction, and noise robustness. Our approach innovatively introduces a multi-band driven dynamic graph construction mechanism to finely capture multi-scale correlations, employs a dual-granularity adaptive graph filter to distinguish and integrate local and global dependencies, and adopts a hierarchical graph reading strategy to enhance robustness against noise and anomalies. Extensive experiments on real-world datasets, including Climate, Electricity, Weather, and PEMS series, demonstrate that FAST-GNN significantly outperforms state-of-the-art baselines in both long- and short-term forecasting tasks, particularly showing stable and superior performance in challenging scenarios such as high-noise environments and long-term horizons.
Shuming Zhang· International Conference on...· 0 citations
The Indian Summer Monsoon shows significant spatial variation. While prior work primarily focused on forecasting rainfall amounts, little attention has been given to how consistently a location's seasonal rainfall trajectory repeats from year to year. We introduce a graph-based machine learning framework to classify locations across India by this inter-annual consistency. Using 2001 to 2022 GSMaP ISRO data (excluding 2012), we constructed graphs for 29,026 grid points where nodes represent individual years and edges denote cosine similarity. A Graph Convolutional Network classified locations as either consistent or erratic with 96.8% accuracy. Applied to the Indian landmass, the model successfully identified the Western Ghats, Northeast India, and parts of central India as consistent regions. This classification was rigorously validated through statistical testing and temporal stability analysis, showing 93.6% agreement across two independent timeframes. Crucially, the results reveal a previously unreported coupling: regions with higher rainfall volumes are also the most temporally repeatable year-to-year, demonstrating an emergent, spatially coherent structure.
Pradyumnan Raghuveeran, Gaurav Chopra, Ajay Bankar et al.· 0 citations
Accurate regional weather forecasting in complex maritime environments is challenging due to intricate atmospheric dynamics. While deep learning presents a promising alternative to numerical weather prediction (NWP), current paradigms face an inherent tradeoff. Autoregressive (AR) models suffer from recursive error accumulation and spectral decay, whereas pure computer vision approaches neglect physical laws, yielding dynamically inconsistent predictions. To address these limitations, we propose the physics-informed non-AR network (PINA-Net) to reconcile visual sharpness with physical consistency. Our framework synergizes a lightweight 3-D spatiotemporal encoder with coordinate attention and a cumulative residual strategy, enabling the one-shot generation of high-fidelity sequences without error propagation. Crucially, we integrate a physics-constrained loss function that embeds partial differential equations (PDEs) for mass conservation and divergence directly into the optimization process. Extensive experiments on a high-resolution meteorological dataset demonstrate that PINA-Net significantly outperforms state-of-the-art baselines. The model achieves superior numerical accuracy and structural similarity while effectively suppressing nonphysical artifacts in wind vector fields. Furthermore, by evaluating predictive uncertainty, the framework provides reliable probabilistic boundaries, offering a robust solution adhering to the intrinsic kinematic constraints of the atmosphere.
Yi Yan, Jiangting Li, Yuxuan Wang et al.· IEEE Transactions on Geoscie...· 0 citations
Climate prediction plays a critical role in environmental monitoring, disaster preparedness, agricultural planning, and sustainable resource management. Existing forecasting approaches primarily focus on either temporal sequence learning or spatial feature extraction, which limits their capability to capture the complex interdependence among climatic variables distributed across geographical regions and time. To overcome this limitation, this study proposes a novel Spatio-Temporal Graph Neural Network (ST-GNN) framework that jointly models spatial relationships and temporal climate dynamics within a unified deep learning architecture. The novelty of the proposed work lies in integrating graph-based spatial representations with temporal neural learning to simultaneously capture geographical connectivity and evolving climatic patterns for multivariate climate forecasting. In addition, the framework introduces a unified spatio-temporal learning mechanism capable of handling interconnected environmental variables efficiently. Experimental evaluation demonstrates the effectiveness of the framework with a Mean Absolute Error (MAE) of 7.6050% and a Root Mean Squared Error (RMSE) of 14.8201%. The model achieved 99.52% classification accuracy with precision, recall, and F1-score values of 98.79%. Multivariate climate data collected from a Kaggle dataset were used for training and evaluation.
R. Lakshmi, P.Selva, A. Chinnasamy· 2026 7th International Confe...· 0 citations
Geomagnetic data are crucial for pre‐earthquake electromagnetic anomaly analysis and deep earth exploration, yet data acquisition is highly susceptible to cultural noise, instrument failures or transmission anomalies, leading to data loss or saturation. For decades, time‐series editing and reference station methods have evolved independently. Time‐series denoising without reference constraints risks over‐ or under‐processing, whereas reference station methods demand strictly high‐quality reference data, lacking flexibility. Integrating their strengths, this paper proposes a reference‐channel‐constrained deep learning method for geomagnetic noise suppression. First, a novel network named GATCN—combining a gated recurrent unit, attention mechanism and temporal convolutional network (TCN)—is proposed for the rapid, high‐precision identification of high‐quality or anomalous signal segments. Second, under the constraint of synchronous reference data, multivariate variational mode decomposition (MVMD) decomposes the identified high‐quality signals into distinct frequency components. Third, a convolutional neural network (CNN)–long short‐term memory network (LSTM) network predicts high‐quality signals for noisy or anomalous segments based on the decomposed components, utilizing the same synchronous reference constraints. Finally, the identified and predicted high‐quality signals are chronologically concatenated to reconstruct the complete data. Validation using data from the Western Pacific seafloor and the Chinese mainland demonstrates that GATCN achieves 99.25% recognition accuracy with robust feature extraction and stable gradient propagation, outperforming state‐of‐the‐art models like ResNet, TCN, transformer and IncepTCN. Moreover, the MVMD and CNN–LSTM‐based prediction approach surpasses advanced networks, such as transformer, CNN–transformer, LSTM and bi‐directional LSTM (BiLSTM). Ultimately, compared to existing time‐series methods, our approach substantially enhances reliability and robustness against varying noise intensities and types; compared to the magnetotelluric remote reference method, it drastically lowers reference data requirements, yielding significantly greater flexibility.
Guang Li, Linfeng Li, Jiayong Yan et al.· Near Surface Geophysics· 0 citations
A hierarchical hybrid spatiotemporal architecture is introduced, which synergistically utilizes dilated causal convolutions to extract local trends and instantaneous fluctuations, while incorporating a multi-head self-attention mechanism to aggregate global context information, thereby achieving a complementary fusion of local and global features.
Peng Chen, Danhong Zhang, Yixin Su· International Journal of Gre...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
