Revealing the mutual information between seismic and geodetic data in Guerrero, Mexico

<p>Significant slow-slip events are primarily observed in geodetic data and may occur during tremors and low-frequency, often referred to as "slow earthquakes" seen in the seismic data. The link between slow earthquakes and slow-slip events is still being discussed nowadays, mainly because of the difficulty of defining the seismic signatures of slow earthquakes on the seismic signal level. However, detecting the seismic signatures associated with slow-slip events would be a crucial step toward understanding these phenomena and their role in the seismic cycle. This study demonstrates how machine learning principles can help interrogate the relationship between geodetic and seismic data. We condense the information carried by the continuous seismograms recorded by a single station in Guerrero, Mexico, over a decade. Using a deep scattering network with ICA, we obtain seismic features at an hourly sampling rate that we compare with the geodetic data. Our method shows that we can explain the geodetic movements with a few seismic features, suggesting that the seismic radiations of slow-slip events are well-defined. We further examine the properties of the associated seismic wavefield and infer the contribution of each class to the overall displacement observed in the geodetic data.</p>