Prediction of Gamma Ray data from pre-stack seismic reflection partial angle stacks using Continuous Wavelet Transform and convolutional neural network approach

A key concern for geophysicists working in the oil and gas industry is reservoir characterization, which is the process of building a qualitative and quantitative description of a reservoir to optimize its lifetime performance (Fayers and Hewett, 1992; Chierici, 1992). For this purpose, various reservoir properties derived from both direct and indirect measurements are utilized to build a full field model and simulate a field performance. Often, the most important indirect measurement of reservoir and fluid properties is seismic data (Ma et al., 2005; Alabi and Enikanselu, 2019). Direct interpretation of seismic data to characterize the reservoir in terms of facies and hydrocarbon presence is a significant challenge owing to its limited spatial and vertical resolution, and there are a wealth of seismic inversion algorithms available and under development which aim to estimate elastic properties such as bulk density (ρ), compressional acoustic velocity (vp), shear acoustic velocity (vs) and volume of shale (Vsh), directly. Until recently, seismic inversions have been either deterministic or stochastic. The results of deterministic seismic inversion tend to be low in resolution and are very sensitive to noise and data quality as these methods are based on physical principles (Aamir et al., 2017). Comparisons between stochastic inversions based on Markov Chain Monte Carlo (MCMC) approaches and deterministic inversions on the estimation of seismic impedance and lithofacies show that each stochastic simulation is able to provide more resolution than a deterministic solution (Li et al., 2018). Stochastic inversion models are mainly based on MCMC modified approaches and implement a wide list of target parameter realizations considering the range of uncertainties (Lang and Grana, 2017; Yang and Zhu, 2017). However, each realization (implementation) from a stochastic seismic inversion is perceptibly different from the others, leading to an ambiguous interpretation of the results. On the other hand, averaging all realizations to simplify the interpretation brings deterioration in the resolution of the output attributes.