Summary Accurately characterizing fine-scale mechanical heterogeneity in shale reservoirs is essential for improving the resolution and physical fidelity of geomechanical models. Conventional well logs typically provide decimeter-scale vertical resolution, which is insufficient to capture the millimeter- to centimeter-scale mechanical variations in laminated shale formations. To address this limitation, we propose a high-resolution geomechanical modeling method integrating scratch testing experiments. Continuous scratch tests were conducted on 11 full-diameter cores from the Wufeng-Longmaxi Formation in a shale gas field of the Sichuan Basin, yielding centimeter-scale profiles of Young’s modulus (YME) and uniaxial compressive strength (UCS). Quantitative correlations between experimental and logging data were established, and a dimensionality-reduction-based fusion algorithm was used to achieve cross-scale precision reconstruction. By combining multisource logging parameters and lithofacies classification, a machine learning surrogate model was developed to predict centimeter-scale YME distributions within the target interval, leading to the construction of a high-resolution in-situ stress model. The results demonstrate that this approach enhances the elastic parameter resolution from the decimeter to the centimeter scale, significantly improving the model’s capability to capture intralayer stiffness contrasts and stress barriers. Incorporating high-resolution parameters revealed more detailed intralayer fluctuations in the maximum and minimum horizontal principal stresses, exposing local stress transitions that conventional models fail to resolve. This study confirms that the multiscale modeling method integrating scratch test experiments enables refined characterization of reservoir stress fields and provides a robust methodological framework for high-resolution in-situ stress characterization, with particular relevance for analyzing vertical mechanical trends and reservoir-scale geomechanical heterogeneity.
Zhang et al. (Thu,) studied this question.