Classification of pre-seismic gravity perturbation and background noises based on scattering network

Abstract

Abstract The surge of tidal gravity data collected in China is outpacing our abilities for analysis, and it is challenging for human expert to deal with such huge datasets. Moreover, there is no chance for experts to detect earthquake related signal in time due to sophisticated signal processing procedure. In this paper, we attempt to use scattering network to extract features from continues gravity data, and apply gaussian mixture model to classify earthquake-related signals and background noises in an unsupervised way. We take The Yangbi and Maduo earthquakes for example. The Yangbi Ms6.4 (99.87°E, 25.67°N) and Maduo Ms7.4 (98.34°E, 34.59°N) Earthquake occurred successively on May 21–22, 2021 in Dali, Yunnan Province and Guoluo, Qinghai Province of China. The 1Hz sampling records from 4 tidal gravimeters near the epicenter, including one superconductivity gravimeters (#066) and three gPhone gravimeters (YL, GRM and LS stations), altogether spanning from 11th May to 22th May, 2021, were obtained and analyzed. Just simple preprocessing for the raw gravity data (earth tide and air pressure corrections), various signals are identified using scattering network, including background noises, coseismic gravity change and possible preseismic gravity perturbation, which could be caused by the slow strike-slip of fault during earthquake preparation. YL station, about 53 km away from the epicenter of the Yangbi earthquake, show no obvious anomaly signal in spectrogram on May 15–19, but the possible gravity perturbation could be detect indirectly using scattering network, same for the other stations. In this paper, we demonstrate how scattering network may be used to identify weak pre-seismic gravity anomaly signals, enabling real-time seismic monitoring with tidal gravimeters array.