Automatic detection of auroral Pc5 geomagnetic pulsation using machine learning approach guided with discrete wavelet transform

Geomagnetic micropulsations are ultra-low frequency (ULF) signals observed in the magnetosphere as well as on the ground. These signals offer us an effective means to study the coupling of the magnetosphere-ionosphere processes in the space field. The most prominent observed type of such ULF waves is the Pc5 pulsations (with a frequency range of 1–7 mHz), known to have their maximum amplitude in the auroral oval. The low magnitude of Pc5 signals withstands against distinguishing them from the background noise. This study presents a machine learning approach for the automatic detection of geomagnetic Pc5 pulsations in the auroral zone using artificial neural networks (ANN) guided by discrete wavelet transform. Our ANN algorithm is validated and tested using a huge amount of datasets of auroral ground-based geomagnetic records from the Svalbard network during the two solar cycles 23 and 24. The wavelet-based coherence was used to determine the signal's consistency detected by the magnetometer stations of the Svalbard network; since they are sensitive to all sorts of space wave-related phenomena that left their footprint on the magnetic field time series. The Daubechies wavelet transform was utilized to classify and extract Pc5 signals from the artificial noise and the results are correlated with the geomagnetic pulsation records as detected by our ANN-based model. The ANN-based model showed a good correlation of an average of 98% for the different phases of the two studied solar cycles. The statistical regression analysis of the post-processed results yielded a high coefficient of determination of R² = 0.9 and norms of residuals of 8–21 nT. The Pc5 events detected by the ANN-based algorithm during the two solar cycles showed a good correlation with the Kp index, which enables our model for space weather forecasting.