Exploring Coronal Heating Using Unsupervised Machine-Learning

The perplexing mystery of how the solar corona maintains itself at a temperature of million K, while the visible disc of the Sun is only at 5800 K has been a long standing problem in solar physics. A recent study by Mondal et al. (2020, ApJ, 895, L39) has provided the first evidence for the presence of ubiquitous impulsive emissions at low radio frequencies from the quiet sun regions. Based on their observed characteristics, the authors find that these emissions, which we refer to as events, meet all of the requirements for being important for coronal heating. From a solar physics perspective, it is very interesting to understand their detailed morphological characteristics. This forms the objective of our study. To put the complexity of the problem in context, we note that in terms of their strength, the weaker features to be studied are about two orders of magnitude weaker than the weakest features reported earlier and are only a few percent of the steady solar emission; in terms of numbers, they occur at a rate of about five hundred events per minute. Based on earlier work with stronger flux densities and theoretical considerations, these features are expected to be compact in the image plane. To characterise the spatial structure of these events, we construct a peak fitting algorithm to find intensity peaks on the Sun, and fit Gaussian or quasi-Gaussian distributions to them. Density-Based Spatial Clustering of Application with Noise (DBSCAN), an unsupervised machine learning algorithm is used to classify the peaks as isolated or clustered. It is also used to obtain robust fits to these peaks, by rejecting noise features, or those that might be contaminated by artefacts from a bright active region present on the Sun. The final objective is to represent the information in the image plane as a set of features that are well-fit with Gaussians. The characteristics of these features can then be further examined to draw conclusions about solar coronal processes. To do a robust statistical analysis, we have applied this tool to a 70 minute dataset with images at every 0.5 seconds at 132 MHz. We present here the preliminary results from our work.

Theme – Machine Learning, Statistics, and Algorithms