GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 3.3 GNGTS 2024 practcal training with a reasonable number of geophones since the number of coefcients in the DCT can be readily adjusted to align with both the data and model requirements. We demonstrated the trained network's capability in generatng data-driven S-velocity model proposals with minimal data misfts between observed and predicted data. Modifcatons in array setngs and source characteristcs may necessitate retraining the network, which, under similar hyperparameters as presented in this work, requires approximately 11.4 efectve hours. The proposed S-velocity model can serve as a startng model for FWI frameworks, ofering the potental to reduce computatonal costs and address the cycle-skipping issue. Acknowledgements We express our grattude to Nicola Bienat for providing valuable assistance and insightul comments that signifcantly enhanced the quality of this work. References Ahmed, N., Natarajan, T., & Rao, K. R. (1974). Discrete cosine transform. IEEE transactons on Computers, 100(1), 90-93. Bohlen, T. (2002). Parallel 3-D viscoelastc fnite diference seismic modelling. Computers & Geosciences, 28(8), 887-899. Garofalo, F., Fot, S., Hollender, F., Bard, P. Y., Cornou, C., Cox, B. R., ... & Yamanaka, H. (2016). InterPACIFIC project: Comparison of invasive and non-invasive methods for seismic site characterizaton. Part I: Intra-comparison of surface wave methods. Soil Dynamics and Earthquake Engineering, 82, 222-240. Moghadas, D., & Vrugt, J. A. (2019). The infuence of geostatstcal prior modeling on the soluton of DCT-based Bayesian inversion: A case study from Chicken Creek catchment. Remote Sensing, 11(13), 1549. Park, C. B., Miller, R. D., & Xia, J. (1999). Multchannel analysis of surface waves. Geophysics, 64(3), 800-808. Wu, X., Shi, Y., Fomel, S., & Liang, L. (2018, October). Convolutonal neural networks for fault interpretaton in seismic images. In SEG Internatonal Expositon and Annual Meetng (pp. SEG-2018). SEG. Corresponding author: felipe.rincon@phd.unipi.it