Prediction of Intensity Variations Associated with Emerging Active Regions using Helioseismic Power Maps and Machine Learning

We developed recurrent neural networks using the long short-term memory (LSTM) architecture to predict the decrease in continuum intensity linked to the emergence of active regions (ARs), before they become visible on the solar surface and start forming sunspots. The goal of this study is to develop a machine learning (ML)-driven interpretable model to predict the starting time and location of emergence that later will be formed into a large AR. The model was trained on observations that included the emergence of 40 ARs and tested on five AR emergence events. The model training was based on observations of continuum intensity, unsigned magnetic flux, and oscillation power maps computed from Solar Dynamics Observatory Helioseismic and Magnetic Imager (HMI) data. For testing the predictive capabilities, the LSTM model uses a time series of the mean oscillation power calculated from Dopplergrams in four frequency ranges and the mean unsigned magnetic flux to predict the time and location of the decrease in the continuum intensity associated with the emerging ARs in a 12 hr time window. The results demonstrate that the LSTM ML analysis of the oscillation power maps can predict the emergence of large ARs several hours before their initial HMI continuum intensity signal becomes visible and at the time when the HMI magnetic flux is at 4%-9.6% of its eventual maximum value, therefore opening perspectives for further development of ML methodology for AR forecasting.