Speaker
Description
The Einstein Telescope (ET) is a third-generation gravitational wave observatory. As a ground-based detector, it is particularly susceptible to seismic noise at low frequencies, particularly for frequencies below 10 Hz. Accurately predicting seismic waveforms can aid in mitigating the impact of seismic noise, thereby enhancing the detector's sensitivity in these frequency ranges—an essential factor for detecting gravitational waves from sources such as supermassive black hole binaries. This study leverages deep learning algorithms due to their capacity to model complex systems and accurately predict three-component seismic waveforms. Our approach involves training a transformer-based model to utilize initial earthquake waves (P-waves) for the prediction of subsequent, more destructive waves, including S-waves and surface waves. The training process employs synthetic seismograms embedded in realistic noise, with the synthetic data generated using realistic source parameters and Green’s function databases derived from a one-dimensional Earth model. The latter was constructed using AxiSEM, a spectral element code for simulating global seismic wave propagation. The proposed model is capable of predicting seismic waveforms from either a single station or an array of seismic stations. The latter is particularly crucial for the ET, as the model can utilize waveforms from seismic stations surrounding the telescope to predict the waveform at the mirror location.
