Understanding Heating Frequency of Active Region Loops through Forward Modeling and Machine Learning

Understanding how loops in active regions are heated is a critical step in solving the coronal heating problem. In particular, constraining the frequency at which individual strands are re-energized can shed light on what mechanism releases energy from the highly-stressed magnetic field into the coronal plasma. To address this problem, we forward model time-dependent AIA intensity maps for active region NOAA 1158 using a combination of loop hydrodynamics, potential field extrapolations derived from HMI magnetograms, and detailed atomic physics. We model the AIA intensity for a range of heating frequencies and constrain the total energy input based on both observed active region flux and the magnetic field strengths derived from the field extrapolation. We then apply the time lag method of Viall and Klimchuk (2012) to compute cross-correlations for all possible channel pairs for every pixel in our synthesized active region. For a given channel pair, the delay which maximizes the cross-correlation provides a proxy for the cooling time between the two channels in a given pixel. We apply this same technique to twelve hours of AIA observations of NOAA 1158. To make meaningful comparisons between our synthetic and observed data, we train a random forest classifier on the synthesized time lags and apply it to our observed timelags in order to classify the heating frequency in each pixel of the active region. This approach allows us to easily and efficiently incorporate every channel pair in deciding which heating model is most consistent with our observed time lags in the context of our model. We also compute emission measure distributions from our modeled and observed intensities using the method of Hannah and Kontar (2012), as any successful heating model should be able to reproduce multiple observational signatures. Furthermore, we apply this analysis to several more active regions from the catalog compiled by Warren et al. (2012). In order to efficiently analyze this large time-dependent, multi-wavelength data, we use the Dask Python library for out-of-core data processing in order to take advantage of multiple computing cores when preparing and analyzing the data. Such an approach provides a pipeline for processing a many hours of full-disk, level 1 images into a series of time lag maps in a matter of a few hours. This novel combination of distributed and parallel data processing, detailed forward modeling, and machine learning allows us to survey active region heating properties at an unprecedented scale.