AmbSoWi-ML

This is a machine learning method to predict the ambient solar wind flows observed in near-Earth space. The input data are variables from solar coronal magnetic models - the flux tube expansion factor and the distance to the coronal hole boundary along with the solar wind speed measured at L1 from one Carrington rotation before. The model is a decision tree method, specifically a Gradient Boosting Regressor (Python-based), trained on data from 1992 till 2006 and tested on data from 2006 till 2017. It has not been implemented to run in real-time.

Does not run in real-time.

The input is a combination of coronal magnetic field variables and solar wind speeds at L1 from the last solar rotation (t-26, -27 and -28 days). The flux tube expansion factor fp and the distance to the coronal hole boundary d were extracted from coronal magnetic field models. These were updated for every available timestep, producing a set of variables every 3.64 hours. The output from multiple coronal model solutions was used: in the final version, fp and d were extracted from 3 different ADAPT realisations.

The solar wind speed near Earth's bow shock is predicted. Since the machine learning model was trained on OMNI data, the exact location will depend on the timestamp and OMNI's bow shock-calculation algorithm. We can assume average bow shock distance for simplification.