With the development of machine learning technology, research cases for spatial estimation through machine learning approach (MLA) in addition to the traditional geostatistical techniques are increasing. MLA has the advantage that spatial estimation is possible without stationary hypotheses of data, but it is possible for the prediction results to ignore spatial autocorrelation. In recent studies, it was considered by using a distance matrix instead of raw coordinates. Although, the performance of spatial estimation could be improved through this approach, the computational complexity of MLA increased rapidly as the number of sample points increased. In this study, we developed a method to reduce the computational complexity of MLA while considering spatial autocorrelation. Principal component analysis is applied to it for extracting spatial features and reducing dimension of inputs. To verify the proposed approach, indicator Kriging was used as a benchmark model, and each performance of MLA was compared when using raw coordinates, distance vector, and spatial features extracted from distance vector as inputs. The proposed approach improved the performance compared to previous MLA and showed similar performance compared with Kriging. We confirmed that extracted features have characteristics of rigid classification in spatial estimation; on this basis, we conclude that the model could improve performance.

This paper develops a reliable deep-learning framework to extract latent features from spatial properties and investigates adaptive surrogate estimation to sequester CO₂ into heterogeneous deep saline aquifers. Our deep-learning architecture includes a deep convolutional autoencoder (DCAE) and a fully-convolutional network to not only reduce computational costs but also to extract dimensionality-reduced features to conserve spatial characteristics. The workflow integrates two different spatial properties within a single convolutional system, and it also achieves accurate reconstruction performance. This approach significantly reduces the number of parameters to 4.3% of the original number required, e.g., the number of three-dimensional spatial properties needed decreases from 44,460 to 1920. The successful dimensionality reduction is accomplished by the DCAE system regarding all inputs as image channels from the initial stage of learning using the fully-convolutional network instead of fully-connected layers. The DCAE reconstructs spatial parameters such as permeability and porosity while conserving their statistical values, i.e., their mean and standard deviation, achieving R-squared values of over 0.972 with a mean absolute percentage error of their mean values of less than 1.79%. The adaptive surrogate model using the latent features extracted by DCAE, well operations, and modeling parameters is able to accurately estimate CO₂ sequestration performances. The model shows R-squared values of over 0.892 for testing data not used in training and validation. The DCAE-based surrogate estimation exploits the reliable integration of various spatial data within the fully-convolutional network and allows us to evaluate flow behavior occurring in a subsurface domain.

Summary: Ensemble-based data assimilation, when combined with generative adversarial networks (GANs), has proved effective for history matching based on production data in petroleum engineering research. However, the geological realizations produced are often constrained by the characteristics of the initial ensemble and require extensive training data sets to achieve diversity. To address these limitations, we introduce a novel method that enhances history matching in reservoir simulations by integrating a geological-style-mixing approach with GAN-based optimization (StyleGAN). This method addresses the challenges of capturing complex geological features and heterogeneities that influence reservoir performance. We use the StyleGAN architecture to generate diverse geological scenarios with enhanced style diversity. By adapting the style-mixing mechanism of StyleGAN for geological modeling, we developed a framework capable of producing a variety of geological styles. Each style possesses unique characteristics that are distinct from those of the initial ensemble. This approach combines the styles from assorted geological realizations to create new realizations that exhibit a broad spectrum of geological features, thereby significantly improving the history-matching process. The effectiveness of our method is demonstrated through case studies involving a 2D binary permeability field, a 2D Gaussian permeability field, and a 3D bimodal log permeability distribution. Our optimized models displayed considerable improvement over conventional GAN-based optimizations. The correlation with the reference model increased from 0.94 to 0.98 for the binary permeability field, from 0.97 to 0.99 for the Gaussian permeability field, and from 0.97 to 0.99 for the 3D bimodal permeability field. In addition, the production rate matching error improved from 66% to 86% for the binary permeability field, from 81% to 93% for the Gaussian permeability field, and from 81% to 88% for the 3D bimodal log permeability field, with substantial reductions in the root mean squared error (RMSE) compared with the initial model. The proposed method was compared with the previously developed convolutional neural network-principal component analysis (CNN-PCA) and demonstrated similar history-matching performance. However, it qualitatively showed better preservation of the geological style of the trained reservoir ensemble. These findings demonstrate that integrating a geological-style-mixing approach with GAN-based optimization presents a promising avenue for overcoming the limitations of current ensemble-based history-matching methods, particularly in scenarios characterized by high geological complexity and data uncertainties. This research advances history-matching methodologies and signifies the potential of machine learning and artificial intelligence in enhancing reservoir simulation and management.

Urban geological hazards have increased rapidly in metropolitan areas, with ground subsidence incidents rising five-fold from 69 cases in 2014 to 338 cases in 2018. This study presents a comprehensive quality analysis of borehole data across Seoul Metropolitan Area to enhance urban geotechnical information reliability for underground space safety. A total of 14,016 unique boreholes from 67,054 original records were systematically processed using data from Korea Public Data Portal (data.go.kr) and digital elevation model from the National Geographic Information Institute (ngii. go.kr). Data preprocessing revealed significant quality issues including layer thickness calculation errors and duplicate records, which were systematically corrected. Geotechnical analysis showed Seoul’s subsurface comprises primarily sedimentary layers (26.5%), landfill layers (19.5%), and weathered rock layers (17.5%). Altitude-based analysis confirmed distinct characteristics. High-altitude areas dominated by weathered soil and rock formations, while low-altitude areas predominantly composed of sedimentary deposits, reflecting Seoul’s complex topography from mountainous terrain to alluvial plains. Correlation analysis between Korean geotechnical classifications and Unified Soil Classification System (USCS) standards demonstrated quantitative relationships. Elevation accuracy assessment revealed 75.5% of boreholes exhibited good accuracy within 10 m difference, while 8.5% showed large errors exceeding 50 m. This comprehensive analysis provides essential baseline data for 3D geological modeling, hazard assessment, and underground development planning, supporting smart city-based integrated geospatial platforms for sustainable urban development.