Enhancement of Bayesian Seismic Inversion Using Machine Learning and Sparse Spike Wavelet: Case Study Norne Field Dataset

Abstract

The concept of uncertainty is fundamental in seismic inversion modeling, as it pertains to the imprecision or lack of certainty inherent in the model’s results. In this work, we present a comprehensive study that integrates machine learning (ML), sensitivity analysis on well data, and the sparse spike wavelet to enhance to quality of Bayesian linearized inversion (BLI). Furthermore, sensitivity analysis was conducted to assess diverse combinations of attributes to facilitate the identification of the optimal model configuration and ensure robust results. Subsequently, the wavelet, obtained through the application of the sparse spike convolution on seismic data, was used to further augment the quality of results in the BLI process inversion. This wavelet yields a more precise and efficient representation of seismic events, enabling an enhanced quality in the interpretation of data in the Bayesian context of seismic information. By integrating the sparse spike wavelet with the well and seismic data, a substantial improvement in inversion quality is achieved compared to the initial inversion performed using the original well data and the Ricker wavelet. The combination of ML, sensitivity analysis, and sparse spike wavelet in the BLI inversion process generates results that are more precise and reliable, facilitating well-informed decision-making in the exploration of underground resources. In our analysis, we applied the BLI inversion process in real data from the Norne Field in the North Sea, Norway.