Work Description

Date: 16 May, 2025

Dataset Title: One-Hour Ahead Global Probabilistic Geomagnetic Perturbation Forecasting using Deep Gaussian Process

Dataset Creators: Chen, H., Chen, Y., Huang, Z., Zou, S., Huan, X., Toth, G.

Dataset Contact: Hongfan Chen [email protected]

Funding: This work is supported by the National Science Foundation (NSF) under grant number PHY-2027555: SWQU: NextGen Space Weather Modeling Framework Using Data, Physics and Uncertainty Quantification.

Key Points:

- GeoDGP is a data-driven model that probabilistically forecasts local geomagnetic perturbation over the globe at 1-minute cadence 1-hour ahead.
- We evaluate the model on a wide range of geomagnetic storms and at over 200 magnetometer stations across the globe.
- GeoDGP outperforms both the state-of-the-art first-principles Geospace model and the data-driven DAGGER model across multiple metrics.

Research Overview:

Accurately predicting the horizontal component of the ground magnetic field perturbation (dBH), which can be used to calculate the Geomagnetically Induced Currents (GICs), is crucial for estimating the space weather impact of geomagnetic disturbances. In this work, we develop a new data-driven model GeoDGP using deep Gaussian process (DGP), which is a Bayesian non-parametric approach. The model provides global probabilistic forecasts of dBH at 1-minute time cadence and with arbitrary spatial resolutions. We evaluate the model comprehensively on a wide range of geomagnetic storms, including the 2024 Gannon extreme storm. The results show that GeoDGP significantly outperforms the state-of-the-art physics-based first-principles Space Weather Modeling Framework (SWMF) Michigan Geospace model and the data-driven DAGGER model.

Methodology:

The dataset includes the data to reproduce the figures and results presented in the related publication below. While the raw measurements are not included in this folder, they can be downloaded from the respective sources: the SuperMAG website for magnetic field perturbation data and the OMNI website for solar wind data. For SuperMAG dataset, please download YYYY One Minute Resolution Data (All Stations) and select subtract baseline.
Additionally, the dataset includes a fine-tuned Geospace simulation for the Gannon storm.

Files contained here:

The folders show divisions based on contents. The data folder contains necessary data used to reproduce figures and results in the paper. The figure folder contains figures used in the paper. The scripts folder contains scripts used to implement the model training, evaluation and visualization. The detailed information is described below:
The data folder: this folder contains all data used in our analysis.
- AllStations_AllYear_1min_raw: processed yearly magnetic perturbation data of magnetometer stations with pre-computed location information in solar magnetic (SM) coordinate system.
- AMPERE: AMPERE (Active Magnetosphere and Planetary Electrodynamics Response Experiment) field-aligned current data.
- DAGGER: predictions of the DAGGER model introduced in Upendran et al. (2022).
- Geospace_Gannon_MAGGRID: a frame that corresponds to the peak time of the 2024 Gannon extreme geomagnetic storm from a fine-tuned Geospace simulation.
- Geospace_Qusai: Geospace simulations conducted in Al Shidi et al. (2022).
- Input: The input data used for model evaluation.
** TestSet3: this folder contains input of GeoDGP for the test set 3.
** Gannon_storm: this folder contains magnetometer station measurements and Geospace station prediction for the test set 3.
** Other files: GeoDGP model input.
- MT_sites: Magnetotelluric (MT) data of two site near magnetometer station OTT.
- omni: an empty folder. Data can be downloaded from the OMNI website using the script.
** high_res_1min: 1-minute resolution OMNI data
** low_res_1h: 1-hour averaged OMNI data.
- test_station: evaluation results.
** TestSet*: results for each test set
** FFT_withUQ: results of fast Fourier transform and transformer heating analysis
- train: training data and the trained model state.
** model: trained model state that can be used to load the model.
** scaler: scalers used for standardization.
The figure folder: this folder contains all figures used in the paper.

Software:

To reproduce the results: the permanently archived release of the code is available at https://doi.org/10.5281/zenodo.15505660.
To keep track of the latest developments: the actively maintained GitHub repository is available at https://github.com/HongfanChen/GeoDGP/.

Related publication(s):

Chen, H., et al. (2025). GeoDGP: One-Hour Ahead Global Probabilistic Geomagnetic Perturbation Forecasting using Deep Gaussian Process. Space Weather.

Use and Access:

This data set is made available under a Attribution-NonCommercial 4.0 International license (CC BY-NC 4.0).

To Cite Data:

Chen, H., Chen, Y., Huang, Z., Zou, S., Huan, X., Toth, G. One-Hour Ahead Global Probabilistic Geomagnetic Perturbation Forecasting using Deep Gaussian Process [Data set], University of Michigan - Deep Blue Data.