Using Gradient Boosting Regressors to forecast the ambient solar wind from coronal magnetic models

In this study we present a method for forecasting the ambient solar wind at L1 from coronal magnetic models. Ambient solar wind flows in interplanetary space determine how solar storms evolve through the heliosphere before reaching Earth, and accurately modelling and forecasting the ambient solar wind flow is therefore imperative to space weather awareness. We describe a novel machine learning approach in which solutions from models of the solar corona based on 12 different ADAPT magnetic maps are used to output the solar wind conditions some days later at the Earth. A feature analysis is carried out to determine which input variables are most important. The results of the forecasting model are compared to observations and existing models for one whole solar cycle in a comprehensive validation analysis. We find that the new model outperforms existing models and 27-day persistence in almost all metrics. The final model discussed here represents an extremely fast, well-validated and open-source approach to the forecasting of ambient solar wind at Earth, and is specifically well-suited for ensemble modelling or for application with other coronal models.