GNGTS 2024 - Atti del 42° Convegno Nazionale

Session 3.3 GNGTS 2024 Filling actve seismic null space with LSTM G. Roncoroni 1 , I. Deiana 2 , E. Forte 1 , M. Pipan 1 1 University of Trieste, Italy 2 Stanford University, California, USA Introducton Post-stack seismic data analysis plays a crucial role in understanding subsurface structures and petrophysical propertes,ofen associated with a peculiar low or high frequency behaviour. Sinha et al. (2005) highlighted the presence of low-frequency shadows in associaton with hydrocarbon reservoirs, emphasising the signifcance of low-frequency informaton in post-stack seismic data for reservoir characterizaton. Moreover, the applicaton of discrete wavelet transform-based mult- resoluton analysis for spectral enhancement in post-stack seismic data was discussed by Camacho- Ramírez et al., 2016,which remarked on the role of frequency analysis in characterising heavy oil reservoirs. Reiser & Bird (2016) presented case studies of broadband quanttatve interpretaton, emphasising the utlisaton of frequency-related informaton for improved target delineaton and estmaton of reservoir propertes from post-stack seismic data. Additonally, Du et al. (2016) addressed the challenges of low signal-to-noise rato and the importance of considering the main frequency and signal-to-noise rato of seismic data for thin beds interpretaton in post-stack seismic data. Furthermore, Karsli et al. (2006) discussed the applicaton of complex-trace analysis for random-noise suppression and temporal resoluton improvement in post-stack seismic data, emphasising the signifcance of frequency-related analysis for enhancing data quality, while Shi et al. (2009) addressed near-surface absorpton compensaton technology and its applicaton in the Daqing Oilfelds, stressing the importance of frequency-related compensaton techniques for improving the resoluton of post-stack seismic data. Moreover, Chopra et al. (2003) discussed high- frequency restoraton of surface seismic data, indicatng the relevance of frequency-related restoraton techniques for enhancing the resoluton of post-stack seismic data.Therefore, post- stack seismic data analysis encompasses various dedicated frequency-related analyses and methodologies, emphasising the signifcance of low and high-frequency informaton for reservoir characterizaton, atribute predicton, noise suppression, and resoluton enhancement. In fact, low and high-frequency extrapolaton from actve seismic data is essental for various applicatons such as imaging, reservoir characterizaton, and monitoring. Classical methods for low- frequency extrapolaton involve techniques such as full-waveform inversion (FWI) and autoregressive (AR) spectral extrapolaton. FWI with extrapolated low-frequency data has been proposed as an efectve method for determining the low-wavenumber components of the model from extrapolated low frequencies (Sun & Demanet, 2020). Additonally, the autoregressive extrapolaton method has been utlised to extend the spectral bandwidth of seismic data, enabling