LiveWire: Horizontal Geoelectric Field Prediction With 1‐hr Lead‐Time Using Multi‐Fidelity Boosted Neural Networks

• Published in: Journal of geophysical research. Machine learning and computation Vol. 1; no. 4
• Main Authors: Hu, A., Camporeale, E., Lucas, G., Berger, T.
• Format: Journal Article
• Language: English
• Published: Wiley 01.12.2024
• Subjects: geoelectric field; machine learning; prediction; solar wind; uncertainty quantification
• ISSN: 2993-5210; 2993-5210
• DOI: 10.1029/2024JH000151

Geomagnetically Induced Currents (GICs) are electrical currents generated by rapid changes in the geomagnetic field during space weather events, posing risks to power grids and pipelines. Traditional approaches predict GICs indirectly by forecasting dB/dt $d\mathbf{B}/dt$, the temporal variation of the geomagnetic field, which is proportional to the induced electric field via Faraday's law. However, current physics‐based models driven by in situ solar wind measurements offer only 10–30 min lead times, insufficient for power grid operators to take mitigating actions. Additionally, grid operators prefer direct forecasts of the geoelectric field, which directly influences GICs, rather than relying on intermediate predictions of dB/dt $\mathbf{d}\mathbf{B}/\mathbf{d}\mathbf{t}$ that require complex and time‐consuming calculations. We present a novel approach that directly forecasts the horizontal geoelectric field Eh $\left({E}_{h}\right)$ with a one‐hour lead time, bypassing dB/dt $\mathbf{d}\mathbf{B}/\mathbf{d}\mathbf{t}$ predictions. Our method combines magnetometer data, magnetotelluric survey data, and solar wind inputs into a new probabilistic multi‐fidelity machine learning technique, ProBoost, resulting in the LiveWire model. Using data from the Boulder Geomagnetic Observatory (BOU) since 2002, we trained and validated LiveWire on the top 50 geoelectric field events during geomagnetic storms. Our results show that LiveWire outperforms both a persistence forecast and the operational Space Weather Modeling Framework (SWMF) by at least 31% and 23%, respectively. This advancement in geoelectric field forecasting promises more accurate GIC predictions, helping enhance the resilience of critical infrastructure to space weather. Plain Language Summary Geomagnetically Induced Currents (GICs) due to rapid changes in Earth's magnetic field during space weather events pose a threat to critical infrastructure such as electric power transmission grids. Traditional predictions rely on physics‐based models, yielding only 10–30 min lead‐time, insufficient for actionable GIC mitigation. Our study introduces a groundbreaking method that directly forecasts the horizontal electric field component Eh $\left({E}_{h}\right)$ 1 hr ahead, bypassing an estimate of the geomagnetic field. Integrating magnetometer, magnetotellurics (MT) survey, and solar wind data, we develop a probabilistic machine learning technique to forecast the maximum value of the horizontal ground electric field Eh $\left({E}_{h}\right)$ up to 1 hr ahead. Our method offers a transformative step toward effective GIC risk mitigation, contributing to the safeguarding of essential services during geomagnetic disturbances. Key Points LiveWire is a new probabilistic model that forecasts the horizontal geoelectric field one‐hour ahead The performance of LiveWire outperforms that of the persistence/Space Weather Modeling Framework (SWMF) model by at least 31%/23% True Skill Statistics (TSS) and Matthews Correlation Coefficient (MCC) reveals positive skills in identifying storm events 1 hr ahead