Understanding Heating in Active Region Cores through Machine Learning II. Classifying Observations

TL;DR

This paper develops a machine learning approach to classify heating frequencies in solar active region cores using observational data, revealing spatial variations and the importance of emission measure slope.

Contribution

It introduces a random forest classification method to determine heating frequencies across active regions based on multiple observational parameters.

Findings

High-frequency heating dominates the core of the active region.

Intermediate frequency heating is prevalent near the periphery.

Emission measure slope is the most important feature for classification.

Abstract

Constraining the frequency of energy deposition in magnetically-closed active region cores requires sophisticated hydrodynamic simulations of the coronal plasma and detailed forward modeling of the optically-thin line-of-sight integrated emission. However, understanding which set of model inputs best matches a set of observations is complicated by the need for any proposed heating model to simultaneously satisfy multiple observable constraints. In this paper, we train a random forest classification model on a set of forward-modeled observable quantities, namely the emission measure slope, the peak temperature of the emission measure distribution, and the time lag and maximum cross-correlation between multiple pairs of AIA channels. We then use our trained model to classify the heating frequency in every pixel of active region NOAA 1158 using the observed emission measure slopes, peak temperatures, time lags, and maximum cross-correlations and are able to map the heating frequency across the entire active region. We find that high-frequency heating dominates in the inner core of the active region while intermediate frequency dominates closer to the periphery of the active region. Additionally, we assess the importance of each observed quantity in our trained classification model and find that the emission measure slope is the dominant feature in deciding with which heating frequency a given pixel is most consistent. The technique presented here offers a very promising and widely applicable method for assessing observations in terms of detailed forward models given an arbitrary number of observable constraints.