Quantitative geophysical prediction of shale TOC based on neural networks: A case study in the southern Sichuan Basin, China

Accurate prediction of total organic carbon ( TOC ) content is critical for evaluating the quality of hydrocarbon source rocks at drilling sites. Conventional well logging data, however, fall short in providing three-dimensional (3D) quantitative assessments of shale TOC , primarily due to their inability to fully capture physical properties that are strongly associated with organic carbon enrichment, such as resistivity and porosity. In view of this limitation, this study introduces a quantitative method for shale TOC prediction based on inverted parameter volumes that are sensitive to TOC . By integrating neural network modeling and waveform indication simulation, the proposed method uses logging parameters that exhibit strong correlations with TOC . This multi-parameter, nonlinear geophysical prediction technique achieves higher accuracy than conventional approaches and provides a means of establishing the relationship between TOC content and geophysical logging parameters for 3D shale TOC evaluation. Correlation analysis between measured TOC values in core samples and logging parameters identifies density, acoustic transit time, porosity, resistivity, potassium content, uranium content, and others as TOC -sensitive parameters. These parameters are then used to constrain the post-stack seismic waveform indication inversion model. Subsequently, a nonlinear mapping between these TOC -sensitive parameters and measured TOC values is established using a neural network resulting in a quantitative TOC prediction model. Application of the developed inversion model across the study area demonstrates strong agreement between predicted organic carbon levels and laboratory measurements, confirming that the proposed method provides an accurate and feasible geophysical approach for quantitative shale TOC prediction.