Acoustic impedance estimation from combined harmonic reconstruction and interval velocity

Abstract

Low-frequency components of reflection seismic data are of paramount importance for acoustic impedance (AI) inversion, but they typically suffer from a poor signal-to-noise ratio. The estimation of the low frequencies of AI can benefit from the combination of a harmonic reconstruction method (based on autoregressive [AR] models) and a seismic-derived interval velocity field. We have developed the construction of a convex cost function that accounts for the velocity field, together with geologic a priori information on AI and its uncertainty, during the AR reconstruction of the low frequencies. The minimization of this function allows one to reconstruct sensible estimates of low-frequency components of the subsurface reflectivity, which lead to an estimation of AI model via a recursive formulation. In particular, the method is suited for an initial and computationally inexpensive assessment of the absolute value of AI even when no well-log data are available. We first tested the method on layered synthetic models, then we analyzed its applicability and limitations on a real marine seismic data set that included tomographic velocity information. Despite a strong trace-to-trace variability in the results, which could partially be mitigated by multitrace inversion, the method demonstrates its capability to highlight lateral variations of AI that cannot be detected when the low frequencies only come from well-log information.

Deep decomposition learning for reflectivity inversion

“…This approach has been exploited in other deep-learning-based seismic inversion techniques such as Kaur et al (2020), Downton et al (2020) and Wu et al (2021a), but exceeds the scope of this paper. Initial low-frequency models can be constructed from smoothed well-log information, smoothed velocity fields, prior geologic information or more complicated bandwidth extension frameworks (e.g., Bianchin et al, 2019;Gholami & Sacchi, 2013;Lesage et al, 2015).…”

“…(2021a), but exceeds the scope of this paper. Initial low‐frequency models can be constructed from smoothed well‐log information, smoothed velocity fields, prior geologic information or more complicated bandwidth extension frameworks (e.g., Bianchin et al., 2019; Gholami & Sacchi, 2013; Lesage et al., 2015).…”

Prediction of Reflection Seismic Low-Frequency Components of Acoustic Impedance Using Deep Learning

“…However, we are still not able to fully recover the low-frequency components due to the absence of the low frequencies in seismic data or low signal-noise ratio (SNR), although this method may accurately predict the location of major reflection coefficients. Bianchin et al [7] proposed a method to predict the broadband acoustic impedance using harmonic reconstruction and interval velocity, which improves the prediction accuracy of the low frequencies in the AI. However, one of the issues in this method is that the interval velocity derived from raw seismic data can be unreliable, and this will make the prediction unstable if the data quality is low.…”

“…In this paper, we use the high frequencies of seismic data to predict their low frequencies using recurrent neural networks (RNNs), which helps to improve the prediction accuracy of AI inversion. Due to the success of the application of deep learning methods in many different fields [14][15][16][17][18][19][20], deep learning methods have gained attention in geophysics, both in the industry and academia [7,10,21]. The inverse problems in geophysics are similar to problems in many other fields, such as face recognition, natural language processing, and self-driving cars, and involve building a predictive model to predict the future using preexisting data.…”

Methodological Development for the Comparison of Multiple Machine Learning Algorithms in the Reconstruction of Well Loggings in the Middle Magdalena Valley Basin in Colombia

“…The almost complete depletion of the conventional resources of the area today forces to explore other types of geological formations in which the data were not taken, however, repeating the drilling process would be technically and economically unfeasible. To solve this type of problem, many computational techniques have been developed over the years, such as parametric-based ones such as multi-component induction well-logging (MCIL) (Wang et al, 2008) and autoregressive models (Bianchin et al, 2019) and lately, those based on artificial intelligence (AI) techniques, where different Machine Learning (ML) algorithms are used, the objective of which is to make the program learn from known information (well logs) and be able to predict the missing information. Among them are multiple regression algorithms and neural networks (NN) that were used in South-West Iran (Eskandari et al, 2004) to reconstruct shear wave velocity from log data.…”