Well logging is a critical tool for reservoir evaluation and fluid identification. However, due to borehole conditions, instrument failure, economic constraints, etc., some types of well logs are occasionally missing or unreliable. Existing logging curve reconstruction methods based on empirical formulas and fully connected deep neural networks (FCDNN) can only consider point-to-point mapping relationships. Recurrently structured neural networks can consider a multipoint correlation, but it is difficult to compute in parallel. To take into account the correlation between log sequences and achieve computational parallelism, we develop a novel deep-learning framework for missing well-log reconstruction based on state-of-the-art transformer architecture. The missing well-log transformer (MWLT) uses a self-attention mechanism instead of a circular recursive structure to model the global dependencies of the inputs and outputs. To use different usage requirements, we design the MWLT in three scales: small, base, and large, by adjusting the parameters in the network. A total of 8609 samples from 209 wells in the Sichuan Basin, China, are used for training and validation, and two additional blind wells are used for testing. The data augmentation strategy with random starting points is implemented to increase the robustness of the model. The results show that our proposed MWLT achieves a significant improvement in accuracy over the conventional Gardner’s equation and data-driven approaches such as FCDNN and bidirectional long short-term memory, on the validation data set and blind test wells. The MWLT-large and MWLT-base have lower prediction errors than MWLT-small but require more training time. Two wells in the Songliao Basin, China, are used to evaluate the cross-regional generalized performance of our method. The generalizability test results demonstrate that density logs reconstructed by MWLT remain the best match to the observed data compared with other methods. The parallelizable MWLT automatically learns the global dependence of the parameters of the subsurface reservoir, enabling an efficient missing well-log reconstruction performance.

Identifying geologic structures from seismic images is critical in understanding the earth’s evolutionary history, assessing geologic disasters, and predicting resource distribution. Recently, the robust capabilities of big data processing and nonlinear modeling offered by deep learning have led to its widespread applications in seismic structure identification. Researchers have used synthetic seismic data with structure labels generated by numerical simulation to train deep-learning models to avoid the extensive effort and resources required to manually label them. However, most existing studies are task oriented and often simplify or even ignore some common and important geologic features during the simulation process, which limits the performance of the trained model on field data sets. To further improve the generality and flexibility of deep-learning models for structure identification on data sets with complex structural features, we develop an improved parametric 3D structure modeling framework for numerical simulation of synthetic seismic data that includes dip, fold, unconformity, channel, cave, salt body, and composite fault structure. Moreover, we revise some simulation solutions to make the simulated geologic structures better resemble the ones found in the field. To the best of our knowledge, our synthetic seismic data set has the richest structural features to date. The results of comparative experiments on multiple synthetic and field data sets reveal that a deep-learning model trained on data sets with limited geologic features is prone to learn biased representations, which causes the model to misclassify unseen geologic structures with similar seismic responses to the target structure. In contrast, the same deep-learning model trained on our data set provides significantly reduced false predictions of structures with similar seismic responses and improves the identification accuracy of data sets with complex geologic features. Although this study only identifies faults and channels, our synthetic seismic data set can be easily adapted to identify other seismic structures.

The properties of borehole formations, such as porosity, permeability, and water saturation, play a crucial role in characterizing and evaluating subsurface reservoirs. Although core sample experiments offer precise measurements, they are time consuming and cost intensive. An alternative method is to use the logging data to construct an empirical model that predicts formation properties, which is widely studied due to its speed and affordability. Nevertheless, because the response of a logging point reflects its surrounding formation, conventional logging methods relying on point-to-point (P2P) mapping perform poorly in complex reservoirs. Furthermore, the resolution of conventional logging is lower compared with imaging logging. To address these limitations, this study presents a novel approach to predict formation properties based on a deep-learning framework using heterogeneous well-logging data. Our neural network framework takes short sequences of conventional logging data and windowed imaging logging data as inputs. The neural network applies 1D convolution to extract features from the conventional logging sequences and 2D convolution to extract features from the resistivity imaging data. Then, these two feature vectors are fused and fed into a multilayer fully connected neural network to predict formation properties. A case study of a carbonate reservoir demonstrates that our method delivers more accurate predictions of formation porosity, permeability, and water saturation than the P2P, sequence-to-point, and image-to-point prediction methods. Moreover, it is expected that our paradigm will serve as a source of inspiration for forthcoming research endeavors aimed at enhancing the accuracy of predicting borehole formation properties in complex reservoirs.

《地球科学》(ISSN 1000-2383;CN42-1874/P) (原刊名《地球科学——中国地质大学学报》ISSN 1000-2383;CN42-1233/P)，创刊于1957年，是由教育部主管、中国地质大学主办的综合性地学学术理论期刊，月刊，2015年改名为《地球科学》,她以反映中国地质大学教学与科研最新的、高水平的基础地质、应用地质、资源与环境地质及地学工程技术研究成果为主要任务。

River confluence hubs are critical nodes for water and sediment regulation in arid basins, yet the evolutionary patterns of sandbar assemblages under the combined influence of intense an- thropogenic intervention and aeolian interference remain poorly understood. This study focuses on the Hotan River confluence at the southern margin of the Tarim Basin, utilizing a sequence of multi-temporal satellite imagery and geomorphic statistical methods to quantitatively analyze the morphological responses of sandbars to upstream water conservancy projects in the Karakash and Yurungkash rivers. Monitoring results reveal that upstream damming has driven a systematic transition in sandbar scale from "discrete-dynamic" to "intensive-large scale". In the Karakash River, the total number of sandbars steadily decreased from 42 in the pre-dam stage (1990) to 27 in the post-dam stage (2000), while the average individual area increased from 0.382 km2 to 0.472 km2. The most representative adjustment was a non-linear leap in the maximum sandbar scale from 1.319 km2 (1994) to 3.758 km2 (2000). Similarly, the Yurungkash River exhibited an instantaneous sedimentary burst following river closure in 2019, with the total sandbar area surging to a historical peak of 15.161 km2 in that year. Significant reductions in flow variability have induced river type transformation and topological reorganization of sandbars, with the proportion of side bars in the Karakash River surging from 45.2% to 77.8%. This evolution is governed by a synergistic "power reduction-aeolian supply-biogenic locking" mechanism. Dam operation significantly diminishes the effective stream power required for channel incision, promoting the physical consolidation of sandbodies in low-energy environments. Simultane- ously, high-frequency aeolian processes during non-flood seasons provide critical physical pre- filling of the dry riverbed, while the regulated low-energy environment triggers a positive feedback of vegetation colonization, ultimately "locking" the geomorphic skeleton. The "reverse accumulation" mode and topological transition logic identified in this study provide essential modern sedimentary analogs for the refined characterization of fluvial reservoir architecture in arid basins. These findings are critical for identifying skeleton sand bodies in ancient fluvial successions and for evaluating subsurface reservoir heterogeneity and sweep efficiency.