Towards Operational Space Weather Nowcasting via Solar Transients and Their Substructure Auto-identifications

Since we are highly reliant on space-based technology, running human explorations in near-Earth space and planning for human missions to our neighbouring planet, e.g., Mars, building nowcasting and forecasting frameworks with the potential to reduce the risk posed by major solar transients, e.g. Coronal mass ejections (CMEs), has become one of the prime interests.  We investigated approximately 20 magnetic and plasma parameters including magnetic and kinetic energy, helicity, plasma beta, proton velocity, density, magnetic field intensity, temperature, and their fluctuations, ratio of alpha to proton number density, observed to expected proton temperature, Alfvén speed, Mach number, total pressure, and entropy to characterize solar wind plasma within magnetic ejecta (MEs), sheaths (SHs) of interplanetary (I) CMEs, and nonICMEs. We rank the features based on their importance in characterizing these solar wind structures and use the best 15 parameters to train a supervised machine learning model to auto-identify these structures in the solar wind stream. The f1-scores in classifying MEs, SHs, and nonICMEs are found as 0.92, 0.88, and 0.86, respectively, with macro accuracy of ~90%. Furthermore, we quantify the uncertainty in classifying the plasma parcels. Finally, we develop a pipeline and web-based tool using the model that takes input of streaming solar wind plasma at 1 au and auto-classifies them in MEs, SHs, and nonICMEs. This tool enables real-time space weather alerts by automatically detecting the presence of ICMEs, including MEs and SHs in the solar wind.