Global Earth Magnetic Field Modeling and Forecasting with Spherical Harmonics Decomposition

Panagiotis Tigas, Téo Bloch, Vishal Upendran, Banafsheh Ferdoushi, Mark C. M. Cheung, Siddha Ganju, Ryan M. McGranaghan, Yarin Gal, Asti BhattarXiv 2021 · OpenReview

TL;DR

This paper introduces a novel Deep Learning model that forecasts global magnetic field perturbations in Spherical Harmonics space, offering a significant improvement in speed and global coverage compared to traditional simulation-based or sparse ground-station methods. The model achieves state-of-the-art performance, enhancing forecasting accuracy by over 14% on real-world data and 24% on simulation data.

What Problem Does This Paper Address?

Accurately modeling and forecasting the Earth's magnetic field, particularly its dynamic changes driven by solar wind activity, is a critical but challenging task. Existing methods either rely on computationally expensive Magnetohydrodynamics (MHD) simulations that are slow and resource-intensive, or they use sparse ground-based magnetometer data (like SuperMAG) which provides limited spatial coverage. This gap makes it difficult to obtain a comprehensive, real-time understanding of geomagnetic disturbances, which can have significant impacts on our technological infrastructure, such as power grids, through geomagnetically induced currents (GICs). The paper aims to bridge this gap by developing a method that provides global coverage at a much higher temporal resolution (one-minute cadence) without sacrificing accuracy.

Methods

The core of the paper's approach lies in two key areas:

  1. Spherical Harmonics Decomposition: The researchers represent the global magnetic field as a set of spherical harmonic coefficients. This mathematical technique allows for a compact and efficient representation of a global field from potentially sparse data. They employ a compressed sensing technique, specifically Lasso regression, to recover these coefficients from sparse measurements. Lasso regularization is crucial here as it promotes sparsity in the coefficients, preventing overfitting and ensuring a more parsimonious model.
  2. Deep Learning Forecasting Model: A Deep Learning architecture, combining a Gated Recurrent Unit (GRU) and a Multilayer Perceptron (MLP), is used to forecast these spherical harmonic coefficients. The GRU processes sequences of solar wind data to create an embedding, which is then fed into the MLP to predict the coefficients. This model operates directly in the spherical harmonics space, bypassing the need for computationally intensive MHD simulations and avoiding manual feature engineering, which was common in prior state-of-the-art models.

Datasets

The study utilizes the following datasets:

  • OMNI Dataset: This dataset provides solar wind data, including interplanetary magnetic field (IMF) components, velocity, and temperature, collected at the L1 Lagrangian point. This data is used as input for the forecasting model.
  • SuperMAG Dataset: This dataset consists of geomagnetic perturbation measurements from approximately 200 ground-based magnetometers globally. It's used for evaluating the reconstruction and forecasting performance of the proposed model.
  • MHD Simulation Data: Data derived from a physics-based Magnetohydrodynamics (MHD) model (Open GGCM) is used as a "ground truth" for evaluating the reconstruction and forecasting capabilities, especially for assessing global coverage and finer resolution performance than SuperMAG alone can provide.

Key Insights and Ideas

  • Spherical Harmonics as a Compact Representation: The paper demonstrates that global magnetic field data, even when derived from sparse measurements or simulations, can be effectively compressed and represented using spherical harmonic coefficients. This is a fundamental insight that enables efficient global modeling.
  • Deep Learning in Spherical Harmonics Space: Moving the forecasting task into the spherical harmonics domain, rather than predicting field values at specific locations, proves to be a highly effective strategy. This allows the model to learn global patterns and relationships more efficiently.
  • Eliminating Feature Engineering: By using raw solar wind data directly with a GRU and MLP, the model avoids the need for manual feature engineering, which can be time-consuming and may miss subtle but important relationships. This suggests a more robust and potentially more generalizable approach.
  • Reconstruction from Sparse Data: The study highlights the power of compressed sensing techniques like Lasso regression in accurately reconstructing global magnetic field maps from limited ground-based observations.

Key Results

  • Improved Forecasting Accuracy: The proposed Deep Learning model outperforms the state-of-the-art empirical model by Weimer et al. (2013). It achieves a 14.53% improvement in RMS error on the SuperMAG dataset and a 24.35% improvement on MHD simulation data.
  • Global Coverage at High Cadence: The model provides global coverage of magnetic field perturbations at a one-minute cadence, a significant improvement over existing methods that are either computationally prohibitive or spatially limited.
  • Reliable Reconstruction: The spherical harmonics reconstruction, even from sparse SuperMAG data, is shown to reliably reconstruct the global magnetic field, as validated against MHD simulations.

Limitations and Future Work

  • MHD Model Limitations: The authors acknowledge that while MHD simulations provide a valuable global ground truth, they may miss many short-scale phenomena that actual measurements would capture.
  • Data Sparsity in Southern Hemisphere: The analysis primarily focuses on the Northern Hemisphere due to its more extensive coverage in the SuperMAG dataset, suggesting potential limitations in fully capturing global dynamics if Southern Hemisphere data were more sparse.
  • Future Work: The paper implies that the developed framework can be extended to forecast other components of the magnetic field and potentially other space weather phenomena. The release of code and models post-publication will facilitate further research and validation.

Summary updated

Abstract

Modeling and forecasting the solar wind-driven global magnetic field perturbations is an open challenge. Current approaches depend on simulations of computationally demanding models like the Magnetohydrodynamics (MHD) model or sampling spatially and temporally through sparse ground-based stations (SuperMAG). In this paper, we develop a Deep Learning model that forecasts in Spherical Harmonics space 2, replacing reliance on MHD models and providing global coverage at one minute cadence, improving over the current state-of-the-art which relies on feature engineering. We evaluate the performance in SuperMAG dataset (improved by 14.53%) and MHD simulations (improved by 24.35%). Additionally, we evaluate the extrapolation performance of the spherical harmonics reconstruction based on sparse ground-based stations (SuperMAG), showing that spherical harmonics can reliably reconstruct the global magnetic field as evaluated on MHD simulation.
BibTeX
@misc{tigas2021global,
  author = {Panagiotis Tigas and Téo Bloch and Vishal Upendran and Banafsheh Ferdoushi and Mark C. M. Cheung and Siddha Ganju and Ryan M. McGranaghan and Yarin Gal and Asti Bhatt},
  title = {{Global Earth Magnetic Field Modeling and Forecasting with Spherical Harmonics Decomposition}},
  year = {2021},
  eprint = {2102.01447},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2102.01447},
}
Figure