In this First Break article from December 2024, experts Tim Seher, Hassan Masoomzadeh, Yong Ren, Fons ten Kroode, Mark Roberts (TGS), Alexander Kritski, Harald Westerdahl, Mark Thompson and Åsmund Sjøen Pedersen (Equinor) use rotational measurements from a new type of ocean bottom node for the attenuation of non-compressional energy.
Introduction
The attenuation of undesired signals is a continuing issue in the processing of ocean bottom node (OBN) seismic data. An example is the presence of non-compressional energy on vertical geophone records (Paffenholz et al. 2006a, 2006b), which limits our ability to jointly process hydrophone and geophone data. This energy is linked to the presence of both shear body waves and scattered surface waves. It has been called interchangeably shear-wave noise, Vz noise (referring to noise on the vertical geophone), or more recently, non-compressional energy, a term that we will adopt here.
Over the years, a variety of processing techniques have been developed to attenuate non-compressional energy in vertical geophone recordings. The most common approach for attenuating non-compressional energy involves a cooperative denoising procedure between the vertical geophone and the hydrophone in different transform domains such as the Tau-p transform domain (Craft and Paffenholz 2007; Poole et al. 2012), the wavelet transform domain (Yu et al. 2011; Peng et al. 2013; Ren et al. 2020), and the curvelet transform domain (Yang et al. 2020; Kumar et al. 2021; Ren et al. 2022; Kumar et al. 2024). Recently, machine learning (ML) has been applied to accelerate non-compressional energy attenuation in vertical geophone recordings (Sun et al. 2023; Seher et al. 2024). However, a successful application of ML-based noise attenuation requires high-quality training data. New rotational sensors on the seafloor (Pedersen et al. 2023; Kritski et al. 2024; Masoomzadeh et al. 2024) are expected to provide appropriate training data for this purpose.
Figure 1 3D curvelet denoising uses the hydrophone record (a) to attenuate converted body waves and surface waves in the vertical geophone record (b). This process decomposes the input geophone data into a signal model (d) and a noise model (c). The figures shown here come from a conventional four-component OBN and were created by averaging traces for a given offset within a 45° azimuth angle range.
In this paper, we first review four solutions for the attenuation of non-compressional energy in OBN recordings, then demonstrate that rotational records from a prototype receiver can be used to improve the denoising process, and finally show how ML can be utilised to reduce the computational cost and the human effort involved in the noise attenuation workflow.
Read the full article here.