Automatic detection of solar flares observed at 45 GHz by the POEMAS telescope

Abstract

Every 11 years, the Sun goes through periods of activity, with the occurrence of many solar flares and coronal mass ejections, both energetic phenomena of magnetic origin. Due to its effects on Earth, the study of solar activity is of paramount importance. POEMAS (POlarization of Millimeter Emission of Solar Activity) is a system of two telescopes, installed at CASLEO (El Leoncito Astronomical Complex) in Argentina, which monitors the Sun at two millimeter wavelengths (corresponding frequencies of 45 and 90 GHz). The objective of this work is to automatically detect solar flares observed by the polarimeter. First it is necessary to eliminate the background noise, caused mainly by instrumental problems, from the light curves of millimeter solar emission. The methodology used to exclude the noise proposed in this work is to use the tendency of time series. The subtraction of this model from the light curves provides the input to automate the detection of solar flares using artificial intelligence techniques. A Neural Network was trained to recognize patterns and analyze a dataset in order to identify solar flares. Previously, a total of 30 flares had been visually identified and analyzed in the POEMAS database between 2011/11/22 and 2013/12/10. The methodology presented here confirmed 87% of these events, moreover the neural network was able to identify at least 9 new events. As the neural network was trained to detect impulsive events (lasting less than 5 min), long duration bursts were not automatically detected, nor were they detected visually due to the background noise of the telescope. Visual inspection of the POEMAS data, when comparing with microwave data from the RSTN, allowed the identification of an additional 10 long-duration solar flares at 45 GHz. We discuss some problems encountered and possible solutions for future work.