Signal Processing for the Identification of Miyake Events

After the potential impact on communication systems of major solar storms was demonstrated by the Carrington Event of 1859 CE, and subsequent solar proton events, it has become important to develop a system that uses the available historical data for the prediction of similar events. Sudden increases (Miyake Events) in the radiocarbon concentration in the atmosphere were identified from single-year measurements on Japanese trees between 774–775 and 993-994 CE. As part of the ECHOES project, we used state-of-the-art signal processing techniques, as well feature extraction from the Δ¹⁴C measurements to identify the best method for predicting when such events occurred. More precisely, we tested 5 methods including Dynamic Time Warping, Euclidean Distance and COSFIRE filters and spectral features such as centroid, kurtosis and flatness. We computed our results in terms of true and false positive rates. Here, we compare the performance of the methods applied to this task by measuring the false positive rate at 75% of the true positive rate. The best results are achieved with the COSFIRE filters, but some of the spectral features also perform comparably well. With this work we show that machine learning and computational methods are suitable for the identification of possible Miyake Events in the past radiocarbon concentration of the atmosphere.