Dynamic Mode Decomposition of the Core Surface Flow Inverted From Geomagnetic Field Models

Continuous satellite measurements of the Earth's magnetic field have advanced the characterization of spatial-temporal variations of the main field over the past two decades. To comprehend the underlying mechanism responsible for the geomagnetic field variations, we develop a novel core surface flow inversion scheme based on physics-informed neural networks. The inversion method can account for the secular variation contributed by the interaction between the core flow and undetectable small-scale magnetic fields. Based on the novel inversion framework, we derive a time-dependent core surface flow model between 2000 and 2022 from the CHAOS-7 core field model. The inverted core flow is then analyzed using the dynamic mode decomposition to extract wave-like fluid motions. By calculating the magnetic secular acceleration contributed by each dynamic mode, we identify that the dynamic modes with period of about 10 and 7 years are responsible for geomagnetic jerks in the Atlantic and Pacific equatorial regions. Over the past two decades, satellites have been continuously monitoring the Earth's magnetic field. The major part of the field comes from the liquid part of the Earth's core. Geomagnetic measurements show quick changes in the field, including sudden shifts known as geomagnetic jerks. These shifts are believed to be linked to specific fluid motions in the Earth's core. Our study aims to better understand these flows and their effects. We use a method involving neural networks to figure out the patterns of flow at the core surface from the satellite data. We then use a technique to separate these flow patterns into simpler wave-like forms. This helps us see how each wave pattern affects changes in the magnetic field. Our findings suggest that wave-like motions with period of about 10 and 7 years caused geomagnetic jerks in the Atlantic and Pacific regions near the equator. A novel core surface flow inversion scheme based on physics-informed neural networks is developedThe inverted flow from the CHAOS-7 model is analyzed using the dynamic mode decomposition to extract wave-like flow patternsGeomagnetic jerks in the Atlantic and Pacific equator are related to two dynamic modes with period about 10 and 7 years