Abstract Details


Name: Abhishek Kumar
Affiliation: Physical Research Laboratory, Ahmedabad
Conference ID: TVS202510302
Title: Coupling Solar Wind Kinetics with Geomagnetic Storm Dynamics
Authors and Co-Authors: Abhishek Kumar
Abstract Type: Contributory Presentation
Abstract: Space weather, driven by solar phenomena such as solar flares, coronal mass ejections (CMEs), and stream interaction regions (SIRs), strongly influences the coupled magnetosphere–ionosphere–thermosphere (MIT) system. Disturbances within this system can disrupt communication and navigation networks (e.g., GPS, NaVIC), impair satellite functionality, and pose risks to ground-based infrastructure. The intensity of geomagnetic storms is commonly quantified using the Disturbance Storm Time (Dst) index. In this work, we model the Dst response to solar wind kinetic energy input as a forced, damped oscillator, drawing an analogy with an LCR circuit. The associated parameters were estimated and physically interpreted. Because estimating the coefficients of the governing differential equation directly from data is a nonlinear challenge, we employed convolutional neural networks (CNNs) to infer these coefficients and link solar wind kinetic energy input with the Dst response. This provides a physics-informed alternative to conventional AI/ML approaches, which are sometimes applied without sufficient grounding in the underlying physical processes. Building on this, we also developed a feature-based artificial neural network (ANN) model for forecasting the Dst index (Integrating with LCR analogy) using a combination of remote solar observations and in-situ measurements. Inputs to the model include solar wind parameters from the Aditya-L1 ASPEX payload, interplanetary magnetic field (IMF) data, and remote indicators of solar activity. This integrated dataset enables the model to capture both precursor signals and direct solar wind drivers of geomagnetic storms. Comparative analyses demonstrate that the ANN consistently outperforms LSTM and CNN architectures (with 64 units), achieving the highest R² values across all test cases. This highlights the effectiveness of feature-based approaches, especially when guided by physical insight. To enhance interpretability, SHAP analysis was applied, identifying solar wind velocity, IMF Bz, and turbulence spectra as the most influential predictors. Based on these results, forecasts can be translated into warning levels that correspond directly to physical processes within the MIT system.