Mineral resources serve as a critical material basis supporting socio-economic development, with their formation and distribution governed by the complex interactions between deep Earth processes and surface environmental conditions. With the ever-growing global demand for mineral resources, traditional mineral resource prediction methods face significant challenges in their application to covered regions, deeply buried deposits, and non-traditional exploration regions. In recent years, the rapid development of big data and artificial intelligence (AI) technologies has provided significant opportunities for mineral resource research, offering transformative tools for mineral prediction and assessment. This paper systematically reviews the theoretical evolution of mineral prediction and explores a new AI- and big data-powered paradigm, which includes an expanded concept of “ore deposits”, multi-system integrated modeling involving the Earth system, metallogenic system, exploration system, and prediction-evaluation system, intelligent integration of geological survey data and long-tail scientific data, and the deep integration of human intelligence (HI) and artificial intelligence (AI). Based on several representative case studies from recent research projects completed by the author’s team, including integrated mineral prediction in covered regions, quantitative prediction of deep mineral resources, and the construction of a global porphyry copper deposit knowledge graph, this paper demonstrates the innovative application of nonlinear theory and AI techniques in addressing key scientific issues related to mineral prediction. Building on this, the paper anticipates that future data-driven and intelligence-integrated research paradigms will fundamentally transform the paradigm of mineral prediction approaches, significantly enhancing their accuracy and efficiency. This shift will accelerate the transformation of Earth science research from traditional, experience-based practices to intelligent and quantitative methodologies, providing essential theoretical and technological support for the next generation of strategic breakthroughs in mineral exploration.

With the arrival of the big data era, the application of big data technology in mineral exploration has become a trend for future development. This study systematically reviews the development of big data-based mineral exploration and integrated information prediction theories, explores the key technologies of big data in mineral prediction, and presents the following main conclusions based on practical case studies: First, big data mineral exploration can effectively address the challenges of increasing data volume and complexity, providing more accurate data interpretation and predictive support. Second, as a technological tool, big data mineral exploration must rely on solid mineral exploration theories, especially integrated information prediction theory. This theory not only provides theoretical support for big data methods but also improves the accuracy and efficiency of mineral resource predictions. Finally, based on integrated information prediction theory and using a convolutional neural network (CNN) model, mineral prediction for the Baiyinchagan Dongshan-Maodeng area in Inner Mongolia was conducted, demonstrating its application potential in mineral resource prediction. The research findings provide valuable references and practical experience for the application and theoretical development of big data in mineral exploration.

Mineral resource prediction is a key area of research in mathematical geoscience, involving the integration of diverse, cross-disciplinary geoscientific datasets. This process involves significant computational complexity and workload, along with semantic inconsistencies, which pose significant challenges to researchers. The emergence of next-generation generative artificial intelligence, particularly large language models (LLMs) and intelligent agents, is catalyzing transformative advances across industries, facilitating the transition toward intelligent mineral prediction. This study proposes a super-agent framework for AI-driven mineral resource forecasting, built upon multimodal large language models (MLLMs) (e.g., DeepSeek, Qianwen) and agent orchestration frameworks. The super-agent framework comprises a management agent and multiple specialized agent groups (geological, geophysical, geochemical, remote sensing), each comprising multiple atomic agents or lightweight agent collectives. Individual agents are capable of accessing and invoking specific tools—including those locally deployed, cloud-based, or dynamically generated—as well as datasets and services. Upon receiving external prediction tasks, the management agent dynamically coordinates the workflow. This involves orchestrating specialized agent groups, atomic agents (e.g., for geochemical map generation), analytical tools (e.g., interpolation algorithms), and relevant data sources, executing tasks either sequentially or in parallel as needed. Using geochemical map generation as a case study, we elucidate the internal mechanisms of agent-model collaboration, enabling one-click generation of numerous geochemical element maps (up to 39 different elements), thereby validating the effectiveness of this AI-driven approach. By seamlessly integrating multimodal LLMs, intelligent agents, and domain-specific mineral prediction workflows, this framework enables zero-code operation with multimodal interaction through natural language (text or voice), and presents a promising approach for establishing a new paradigm in intelligent mineral resource prediction.

Mineral resource prediction and evaluation have long been important application areas for big data and artificial intelligence (AI) in geosciences research. Since the 1960s, the waves of AI research and technological revolutions have profoundly impacted the development of mineral resource prediction, leading to significant breakthroughs in both theory and technical methods in this field, thereby supporting mineral exploration efforts. In the context of current Earth system science research, quantitative mineral resource prediction needs to move beyond the traditional thinking of “static spatial correlation analysis of ore-controlling factors” and consider the deep-time dynamic evolution history of metallogenic systems, including the “source-transport-storage-transformation-preservation” processes. This requires extending to “multi-factor, cross-scale dynamic integrated prediction” within Earth system science, and developing theories and methods for intelligent mineral resource prediction and evaluation that couple spatiotemporal data. Porphyry deposits are major sources of copper, molybdenum, gold, and other minerals globally, recording key information about the Earth’s lithospheric interactions and material cycles driven by plate tectonics. Both porphyry deposit exploration and plate tectonic reconstruction have accumulated vast amounts of relevant global geospatial data. This paper primarily introduces a data-driven spatiotemporal machine learning model for predicting porphyry deposits, including the construction of deep-time geoscience datasets, machine learning models, and their application cases. We propose a spatiotemporal machine learning model for porphyry deposit prediction that integrates deep-time igneous geochemical data with dynamics parameters of plate subduction. The big data mining revealed that the carbonate subduction flux is a key parameter determining the mineralization potential of arc magmas, providing geodynamic evidence for the critical role of sediment subduction in magmatic-hydrothermal porphyry mineralization. The study also quantitatively evaluates the spatiotemporal distribution patterns and resource potential of porphyry copper deposits in the Andes belt across different geological periods. Therefore, the development of spatiotemporal coupled metallogenic prediction theories and methods can provide important insights and unique perspectives for understanding material recycling and resource effects in deep time, predicting the spatiotemporal distribution of potential mineral resources, thereby guiding mineral exploration.

With the rapid advancement of AI and big data, machine learning-based mineral prospectivity mapping has become a research hotspot. However, models with deeply nested, nonlinear structures and abstract representations often exhibit opaque “black-box” characteristics. This lack of interpretability between predictions and metallogenic processes limits model generalization and reliability. To address this, we propose a knowledge-data driven interpretable prediction method. First, the Best-Worst Method(BWM) is used to derive geological feature weights for constructing an ensemble model, enhancing its performance. A multi-scale, multi-dimensional interpretability framework spanning from global to local levels and from feature-level to sample-level interpretations is then applied to deconstruct results and evaluate feature importance. Expert-guided corrections, informed by field validation, further refine the predictions, thereby forming a closed loop of knowledge embedding and discovery. This process improves workflow transparency and result reliability. Applied to the Keeryin area in Sichuan, the method identified eight high-potential Class A targets, occupying only 6.58% of the study area yet containing 84% of the known deposits. Key predictive features included remote sensing spectral response of albite/albite spectral signature, total alkali (Na₂O+K₂O) content, ring structures, Li/La ratio, and two-mica granite—as validated by their strong spatial correlation with known pegmatite-type lithium deposits in the field.

In the era of big data, machine learning-based intelligent mineral prediction methods have been widely applied. The integration of non-linear theory and techniques, such as fractal and multifractal approaches, into intelligent mineral prospecting research provides new perspectives and technical support. This study focuses on the Habahe gold deposit in Xinjiang, China, and establishes a four-factor information prospecting model based on regional structure, mineralization alteration, magnetic anomalies, and induced polarization anomalies. An intelligent prediction workflow is implemented, by combining the multifractal method with the random forest algorithm. The S-A multifractal filtering technique and local singularity analysis are employed to separate the background variations of regional geophysical and geochemical data from superimposed anomalies, enabling the extraction of concealed information indicative of deep mineralization. The C-Nsum multifractal model is applied to reveal the hidden nonlinear characteristics of gold content in drilling data and determine anomaly thresholds. Subsequently, the random forest algorithm and SHAP method is utilized for comprehensive information integration and feature contribution evaluation, achieving quantitative prediction of gold mineral resources. This approach delineated three prospective mineralization targets, which were validated through drilling, demonstrating the effectiveness of multifractal theory in quantitative mineral prediction within the Habahe gold deposit area. The results provide a robust basis for subsequent mineral exploration efforts.

It has rich lead-zinc mineral resources in Northwest Guizhou. Due to the deep burial of ore bodies, it is difficult to prospecting. Data-driven mineral prospectivity prediction using machine learning (ML) is becoming a powerful tool for exploring deep hidden lead-zinc deposits. However, ML-based prospectivity prediction faces several common issues, particularly insufficient training samples and class imbalance caused by the scarcity of mineralized samples. To address these problems, this paper proposes a K-means clustering-improved conditional tabular generative adversarial network (KC-CTGAN) method for mineralized sample augmentation. Specifically, the density is first judged according to the Euclidean distance between samples in each cluster after K-mean clustering, and expanding more samples in the sparse clusters to increase their density to realize the expansion of the mineralized sample set. Then, the adversarial network generates (GAN) generates new category labels with high abstraction and uses the new category labels for conditional generation, thus improving the quality of augmented samples. Finally, the augmented positive samples and randomly undersampled negative samples are used to construct a sufficiently large and balanced labeled datasets to train a Category Boosting (CatBoost) classifier, and establish a mineral prospectivity prediction model based on KC-CTGAN-CatBoost. The performance of the proposed model was verified by using comparative tests and such as accuracy, recall, precision, F1-score. Experimental results demonstrate that compared to the prediction model constructed without KC-CTGAN-based sample augmentation, the proposed model achieves improvements of 8.7%, 7.4%, 10.2%, and 8.8% in accuracy, recall, precision, and F1-score, respectively, proving the effectiveness of the KC-CTGAN augmentation method in enhancing the performance of the mineral prospectivity prediction model. The prediction results will provide more precise target areas for the exploration of deep-seated concealed lead-zinc ore bodies.

Deep learning models have been widely applied in mineral prospectivity mapping (MPM) due to their powerful ability to extract features from data. However, supervised deep learning methods often face challenges such as insufficient training samples and class imbalance between positive and negative samples. The inherent rarity of mineralization events further compromises model robustness and generalization ability. To address these issues, this study employs three distinct data augmentation methods:1. Sliding Window Augmentation: This method uses known positive and negative samples as centers and performs multiple sliding operations to generate augmented samples; 2. Generative Adversarial Network (GAN) Augmentation: Generative models, specifically GANs, are utilized. The networks are trained on real samples, and augmentation is achieved using the trained generator; 3. Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) Augmentation: Similarly, the WGAN-GP framework is trained on real samples, and its trained generator is used for sample augmentation. These three data augmentation methods effectively expand the sample size while maximally preserving the geological significance of the samples. To validate the effectiveness of augmentation, this study employs the Fréchet Inception Distance (FID) between real and generated samples alongside evaluation using a Convolutional Neural Network (CNN). The results demonstrate that the CNN model trained on the WGAN-GP-augmented dataset exhibits superior generalization ability. Furthermore, the mineral prospectivity map for gold deposits generated for the Gannan area provides important insights for future mineral resource exploration efforts.

Exploration for hidden ore bodies is increasingly important and demands innovative prospecting methods. Data-driven metallogenic prediction models using ensemble learning are becoming powerful tools for deep mineral exploration. However, such models face challenges, particularly in time-consuming hyperparameter tuning requiring extensive computation and expertise. To address this, we propose a framework integrating multi-source geological knowledge with Bayesian Optimization (BO) for ensemble learning. Specifically, a manganese (Mn) metallogenic prediction database integrating multi-source geological knowledge is first constructed. Metallogenic prediction models for Mn ore in northeastern Guizhou are then established using Adaptive Boosting (AdaBoost) and Random Forest (RF). The hyperparameters of these base models are optimized using Bayesian Optimization (BO) via 5-fold cross-validation, resulting in the optimized BO-AdaBoost and BO-RF models. Model performance is evaluated using metrics including accuracy, precision, recall, F₁-score, kappa, and AUC values. Results show significant improvements in AUC for both BO-optimized models compared to their non-optimized counterparts, demonstrating BO’s effectiveness for ensemble learning hyperparameter tuning. Furthermore, the BO-AdaBoost model achieves higher prediction accuracy (92.8%) and generalization performance than the BO-RF model (89.9%), highlighting its strong potential for metallogenic prediction. The prospectivity map generated by the BO-AdaBoost model provides critical guidance for exploring deep-hidden Mn deposits in northeastern Guizhou and can direct future mineral exploration and development.

The Jiaoxibei gold concentration area is dominated by altered rock-type gold deposits, which are strictly controlled by fault structures. The NE-NNE-trending Sanshandao, Jiaojia, and Zhaoyuan-Pingdu fracture zones collectively control over 80% of the gold resources in the Jiaodong Peninsula, making the study of structural ore-controlling mechanisms a primary focus for geologists. Utilizing extensive exploration data from the Jiaoxibei area, this study analyzes and synthesizes the coupling relationship between fracture morphology variations and orebody spatial distribution, thereby revealing key structural controls on mineralization. Altered rock-type gold ore bodies are strictly governed by regional NE-NNE-trending brittle-ductile structures, predominantly occurring beneath the main fracture plane. The spatial distribution of ore bodies is constrained by the activity of ore-controlling structures during the mineralization period. Furthermore, pre-mineralization wave-like structures exert significant control over the spatial morphology and enrichment of the ore bodies. Typical ore bodies exhibit distinct spatial distribution patterns characterized by leaning and slanting arrangements, indicating their formation within a unified tectonic stress field. A marked positive correlation (correlation coefficient = 0.73) exists between the thickness of the alteration zone and the mineralization rate of the ore-controlling structure, reflecting the close association between alteration scale and gold mineralization. Studies of representative altered rock-type deposits in Jiaoxibei reveal that the mineralization intensity of main ore bodies within enrichment zones alternates between strong and weak (or barren) segments along the direction of inclination. This pattern is termed the directional zoning of enrichment along the inclination. The insights gained into these regularities provide a significant basis for predicting the location of deep ore bodies.

The Jiaodong gold deposit is a major concentration area of gold resources in eastern China, characterized by complex geological information and an extensive knowledge system. Traditional information retrieval methods struggle to meet the advanced demands of semantic understanding and knowledge reasoning in mineral exploration. To enhance geological knowledge service efficiency, this study develops an intelligent question-answering system for the Jiaodong gold deposit domain based on GraphRAG (Graph-enhanced Retrieval-Augmented Generation) technology. The research utilizes academic papers from CNKI as the corpus, employs OCR and large language models (LLMs) for text parsing and semantic standardization to establish an ontological knowledge system covering core concepts such as mineralization types, ore-controlling structures, and mineral assemblages. The system uses prompt engineering-driven LLMs to automatically extract entities and relationships, constructing a structured knowledge graph integrated into Neo4j. Furthermore, by combining semantic embedding with community clustering algorithms, a knowledge indexing network enables natural language question answering, semantic query expansion, and knowledge provenance. Evaluation results demonstrate the system’s superiority over traditional RAG and general models (e.g., ChatGPT-4o) in answer accuracy, contextual precision, and knowledge interpretability, exhibiting enhanced professional adaptability and reasoning capabilities. The findings provide a novel technical pathway for intelligent information services in gold deposits and theoretical support for knowledge of graph-enhanced language models in geoscience knowledge management.

The Jiaoxibei gold ore concentration area is characterized by abundant gold resources and diverse metallogenic types. In recent years, significant progress has been made in gold metallogenic theory and deep mineral exploration in this region, driven forward by the continued advancement of national strategic initiatives aimed at achieving breakthroughs in mineral exploration. Based on a comprehensive review of research achievements related to gold deposits in Jiaoxibei, this paper outlines the regional geological setting, analyzes the spatial distribution and metallogenic features of representative deposits, and establishes a regional metallogenic model. Particular attention is given to the genetic processes of gold mineralization, including a systematic summary of the formation mechanisms and differences among major deposit types, as well as a detailed discussion of the coupling interactions among tectonic activity, magmatic intrusion, and hydrothermal fluid migration. Additionally, this study reviews recent advances in gold exploration technologies and methodologies, emphasizing the integrated application of geophysical, geochemical, and remote sensing techniques. The emergence of data-driven prospecting methods is also discussed. Nevertheless, several challenges remain in current exploration practices, including insufficient integration of multi-source data, reliance on traditional geological survey-based models, and the early-stage development of intelligent prospecting technologies. Finally, the paper proposes key future research directions for data-driven mineral exploration in the Jiaoxibei region, aiming to enhance exploration efficiency and promote methodological innovation.

High-precision 3D geological modeling serves as a crucial foundation for the rapid advancement of digital twin technology, providing essential support for resource prediction, engineering planning, and disaster prevention. Traditional 3D geological modeling methods often rely on manual interaction, struggling to meet the demands of precise structural representation and real-time updates in complex geological conditions. To overcome these limitations, the recent introduction of machine learning and deep learning approaches offers new intelligent solutions, significantly improving model automation and the representation of complex geological structures. This paper systematically reviews the development of 3D geological modeling, summarizing technical characteristics across three distinct stages: semi-intelligent modeling, machine learning-based modeling, and deep learning-based modeling. Furthermore, we comprehensively analyze the integrated applications of deep learning with uncertainty analysis, transfer learning, principal component analysis and multiple-point geostatistics. Considering existing challenges such as sparse data processing, computational complexity, model interpretability, and real-time updating capabilities, we propose future research trends, including multimodal data fusion, embedding of geological knowledge, lightweight model optimization, uncertainty quantification and Large Language Models. With ongoing progress in intelligent modeling techniques, the accuracy, reliability, and adaptability of 3D geological models are expected to continuously improve, further advancing the application and engineering practice of digital twin technology in geology.

The Eastern Tianshan region is an important metallogenic belt and exhibits a complex tectonic evolution, with extensive exposures of intermediate-acidic intrusive rocks primarily formed during the Late Paleozoic. Understanding their relationship with regional tectonic evolution and the formation of magmatic-hydrothermal associated metal deposits is of great significance for comprehending the regional tectonic environment and ore-forming patterns. However, the covering layers have resulted in incomplete geological mapping of the intermediate-acidic intrusive rocks in the covered areas of the Eastern Tianshan region. This has hindered our understanding of the regional tectonics and ore-forming patterns there. In recent years, a new paradigm has emerged that integrates multisource survey data, such as geophysics, geochemistry, and remote sensing imagery, using big data analytical techniques to support lithological mapping. Machine learning algorithms have been demonstrated to be powerful tools for data fusion, making them applicable to problems involving the classification and discrimination of complex nonlinear geological data. Therefore, this study proposes using machine learning methods to integrate gravity, aeromagnetic, geochemistry, and remote sensing imagery data to conduct rapid, cost-effective, and more accurate lithological mapping of intermediate-acidic intrusive rocks in the Eastern Tianshan district. In this contribution, the exposed intermediate-acidic intrusive rocks of the study area are labeled as target variables. Furthermore, as predictive variables, Bouguer gravity, aeromagnetic, stream sediment geochemical, and Landsat satellite imagery data are employed. Synthetic minority oversampling technique is utilized to address the issue of imbalanced lithological sample data distribution. Random forest (RF) and artificial neural network (ANN) algorithms are applied, and hyperparameter tuning is conducted through grid search to obtain the optimal prediction models. These models are then used to identify concealed intermediate-acidic intrusive rocks in the covered areas of the Eastern Tianshan region. The results of RF are compared and analyzed with those of ANN. Accuracy, recall rate, and F1 scores indicate that the RF model outperforms the ANN model. Therefore, the prediction results of the RF model are selected as the final result for lithological mapping of intermediate-acidic intrusive rocks in the covered areas of the Eastern Tianshan region. Further discussions are conducted on the control patterns of the spatial distribution of intermediate-acidic intrusive rocks on regional tectonics and mineralization. Compared to traditional geological mapping methods, the machine learning-based lithological mapping approach, which integrates multiple data sources, offers advantages including increased depth, high recognition efficiency, and lower costs, making it an effective method for comprehensively exploring potential geological features and patterns.

For regions with complex terrain, remote sensing technology can be utilized for mineral resource exploration. Hyperspectral remote sensing technology, characterized by high spectral resolution and rich spatial information, enables effective identification of geological features and mineralization anomalies in these areas, and has been widely applied in geological mapping and mineral prospecting. With the high-quality development of China’s aerospace industry, GF-5 hyperspectral data have been employed for the precise identification of alteration minerals. By integrating multi-source information, this approach facilitates the evaluation of prospecting potential in target regions. This study focuses on the area surrounding the Haoyao’erhudong gold deposit in Xinhure area, Inner Mongolia Autonomous Region. GF-5 hyperspectral data were used for alteration mineral information extraction and analysis. Two inter-class distance methods were adopted to select reference endmember spectra, and the traditional Spectral Angle Mapper (SAM) method was applied to complete alteration information mapping. A comprehensive analysis of lithology-structure-alteration multi-source geoscientific elements was conducted to generate an integrated map of alteration minerals. The study successfully extracted eight alteration minerals: kaolinite, muscovite, montmorillonite, alunite, hematite, ilmenite, goethite, and calcite, with corresponding distribution maps created. Comparative analysis between remotely sensed alteration information and existing typical mining zones in the study area validated the applicability of remote sensing-derived alteration data in geological prospecting. In conclusion, multi-source data integration demonstrates enhanced economic practicality in alteration mineral mapping, providing critical guidance and references for identifying potential mineralization zones.

Constructing the metro system is one of the effective solutions to relieve traffic congestion in big cities and enhance the comprehensive carrying capacity and development resilience. High-precision three-dimensional (3D) geological model is an important data infrastructure for determining the geological structure and distribution of unfavorable geological bodies underground. And it is one of the keys to ensuring the safety of metro engineering construction. However, the characteristic that the overall amount of geological data of metro engineering is not large but the local density highly restricts the effective identification and reconstruction of the distribution patterns of geological bodies. Taking the geological structure at a station of Guangzhou Metro Line 11 as the concrete example, this study systematically compares the performance of three modeling methods: random forest (RF), XGBoost and hybrid method of deep learning and multi-point statistics (DL+MPS), under the complex geological conditions of Cretaceous strata, Quaternary sedimentary strata and subvolcanic rocks. The results illustrate that combining the advantages of simulating global features by deep neural network and the local optimization of MPS, the DL+MPS method shows the best performance in key indicators such as accuracy (99.16%), F1 score (98.91%) and AUC value of ROC curve (0.93-0.99). The DL+MPS method can accurately reconstruct the spatial relationship between fault fracture zone and igneous rock mass, and avoid abnormal extension of strata and geological semantic disorder. In contrast, although RF and XGBoost show high training accuracy in the model fitting stage with the accuracy rates of 99.60% and 98.64% respectively, some problems such as discrete distribution of geological bodies, unreasonable extrapolation and strata interpenetration appear in their simulation results. The minimum borehole dispersion value of the results by RF and XGBoost methods reaches 69.93%, is which is significantly lower than that of DL + MPS method (73.33%-87.50%). The research shows that the deep learning model can effectively deal with the challenge of uneven spatial distribution of subway engineering data by virtue of its strong nonlinear feature extraction ability, which provides a better solution for 3D modeling under complex geological conditions, and has important practical value for improving the safety of underground engineering and the application of digital twin system.

Heat flow is a critical parameter in geodynamic studies and resource exploration, but its measurements are often susceptible to interference from factors such as climate and hydrothermal activity, resulting in scarce observational data. This study addresses the insufficient consideration of oceanic crust specificity in existing machine learning-based heat flow prediction models. By integrating measured heat flow and multi-source geological and geophysical data from the South China Sea, we employ Linear Model, Support Vector Machine, and XGBoost algorithms to compare prediction models with and without oceanic crust features (distance to mid-ocean ridge and oceanic crust age), revealing the influence mechanisms of oceanic crust characteristics on the regional heat flow distribution. The results show that oceanic crust features exhibit no significant correlation with measured heat flow values, yet they result in a more pronounced zonal distribution of predicted heat flow near the mid-ocean ridge. In models incorporating these features, the feature importance of Bouguer gravity anomaly becomes significantly higher than that of other features. Furthermore, K-means clustering analysis based on heat flow values, Bouguer gravity anomaly, and distance to mid-ocean ridge identifies two distinct oceanic crust types: tectonically dominated (Type 1) and magmatically dominated (Type 2) regions. This supports the evolutionary trend of the South China Sea’s oceanic crust spreading mechanism shifting from magmatically to tectonically dominated after the mid-ocean ridge jump at 23.6 Ma. This study provides a new data-driven framework for understanding the deep dynamic processes of the South China Sea.

The research in the field of geology is undergoing profound transformations, with the construction of a new knowledge system as its core and big data serving as the driving force. The construction of geological knowledge graphs can effectively address the challenge of knowledge discovery and limited reasoning in scenarios characterized by fragmented data. As one of the critical technologies for constructing knowledge graphs, relation extraction technology plays a pivotal role in identifying relationships between geological entities. Traditional relation extraction techniques are intrinsically contingent upon extensive large-scale annotated datasets. However, the intricacy and specificity of entity relationships in the geological domain render manual annotation of data laborious and time-consuming, consequently leading to a paucity of large-scale labeled datasets. Therefore, the effective implementation of traditional relation extraction techniques within the geological domain is significantly circumscribed. Given the above dilemmas, this study proposes a few-shot learning method for geological relation extraction based on the prototypical network, which innovatively introduces an enhanced prompt learning mechanism and optimizes the instance representation and relation description representation through contrastive learning, thereby significantly improving the representativeness of the prototype. Meanwhile, the weighted loss function and difficult task-assisted training strategy are adopted to enhance the model’s focus on difficult tasks, which effectively improves the overall accuracy. The experimental findings demonstrate that our approach achieves an accuracy of 82.16% in the 5-way 1-shot scenario of a geological few-shot relation extracted dataset. This represents an enhancement of 1.94% over the advanced general-domain model, SimpleFSRE, and 9.01% over the prototypical network, Proto-BERT method. These results substantiate the efficacy of our method.

Rare earth elements (REEs) are critical for modern high-tech industries, yet the metallogenic characteristics and mechanisms of carbonatite-hosted REE deposits remain poorly understood, significantly hindering exploration breakthroughs. With the advent of the geoscience big data era, knowledge graph technology has emerged as a key tool for mineral resource prediction. This study integrates natural language processing (NLP) techniques and knowledge graph construction methods to systematically investigate the metallogenic characteristics and mechanisms of carbonatite-hosted REE deposits. We collected literature pertaining to the Bayan Obo deposit and the Mianning-Dechang metallogenic belt. Using the BERT-BiLSTM-CRF model for entity recognition and the BERT model for relationship extraction, we constructed a domain-specific knowledge graph. Results indicate that minerals, rocks, and elements are critical nodes influencing mineralization. Fluorite exhibits high consistency across regional knowledge graphs, highlighting its potential as a prospecting indicator mineral. Europium and cerium, due to their redox-sensitive anomalies, serve as important indicators for REE exploration. Calcite and bastnaesite also demonstrate indicative potential. The knowledge graph reveals that Bayan Obo exhibits stronger associations with carbonatites, while the Mianning-Dechang belt is more closely linked to alkaline rocks. Hierarchical clustering further demonstrates significant correlations among similar nodes, providing key insights into the metallogenic environment and critical elements. This study offers a novel perspective and methodology for understanding metallogenic mechanisms and provides robust scientific support for the exploration and evaluation of REE deposits.

Copper is a metal resource with high external dependence in China, and porphyry copper deposits represent one of the most critical copper deposit types. To systematically analyze the research status, hotspots, and frontier trends in porphyry copper deposit prediction, this study utilizes literature samples from the CNKI (China National Knowledge Infrastructure) and Web of Science (WoS) databases spanning 1980-2024. Knowledge graph construction and data mining were conducted using CiteSpace and VOSviewer. Through multidimensional analyses of national publication outputs, author-institutional collaborations, and keyword evolution, the results reveal: (1) Iran and China are the most active contributors globally, accounting for approximately 50% of total publications, with Chengdu University of Technology and China University of Geosciences (Beijing) emerging as core research institutions in China. (2) Author collaboration networks and co-citation analyses indicate that core author groups remain underdeveloped both domestically and internationally, yet cross-regional collaborative networks have demonstrated clustering effects, driving research toward systematization. (3) Keyword clustering identifies 11 knowledge modules, while burst detection and visualization analyses highlight “metallogenic conditions and regularities, geological characteristics, and geochemistry” as mature research pillars, whereas “machine learning” and “knowledge graph” represent emerging technological frontiers. The constructed domain-specific knowledge graph provides a panoramic framework for understanding porphyry copper deposit prediction and offers theoretical insights for deep mineral exploration innovation and strategic decision-making.

As a key parameter characterizing the lithospheric thermal state, the spatial distribution of the geothermal gradient is crucial for understanding cratonic thermal evolution and guiding geothermal exploration. Previous studies on the North China Craton (NCC) were largely limited to one-dimensional analyses, failing to resolve vertical variations and limiting model accuracy. This study innovatively develops a depth-stratified prediction model for the NCC geothermal gradient, systematically elucidating its depth-dependent patterns and thermo-tectonic controls. Integrating 573 geothermal gradient measurements with depth information from global databases and literature, data were partitioned into six depth intervals: <500 m, 500-1000 m, 1000-2000 m, 2000-3000 m, 3000-4000 m, and >4000 m. Thirteen geological/geophysical predictors (e.g., Moho depth, geothermal heat flow) were used to train machine learning regression models for each interval. Key results show: (1) Model performance (R²) is depth-dependent: exceeding 0.45 in middle and shallower intervals (<3000 m), but declining significantly at greater depths due to sparse data; (2) Feature importance analysis reveals Moho depth and heat flow dominate deep predictions (weight>40%), while magnetic anomalies and geological age exert minor influence (<10%); (3) The 3D geothermal structure shows systematic depth variations: uppermost high-value zones (>35 ℃/km) align with active fault systems, while mid-deep high-value zones (500-3000 m) migrate eastward, correlating spatially with crustal thinning zones associated with Pacific Plate subduction. This work presents the first 3D geothermal gradient model of the NCC. Results provide critical data for regional geothermal assessment and new constraints on thermo-tectonic coupling during craton destruction. The depth-stratified framework offers a methodological paradigm for thermal studies of analogous tectonic units.

The multidimensional development of Earth’s complex giant system research continues to drive the breakthrough innovation of geoscientific research methods. As an emerging research tool, generative AI technology provides new ideas for geoscientific research by virtue of its powerful data processing and knowledge reasoning capabilities. In this paper, we systematically sort out the development of generative AI technology, compare the technical routes of ChatGPT and the domestic DeepSeek model, and show the advantages of generative AI technology’s powerful natural language processing capability and multimodal model construction. This paper summarizes in detail the application of generative AI technology in geosciences, especially the current status and development trend of its application in the field of resource exploration. Generative AI technology has realized the universal integration of heterogeneous data from multiple sources in the field of geosciences, and has been deeply involved in various stages of work in the field of resource exploration and provided technical support at the levels of data integration, cognitive reasoning, application services, etc., which has set off a wave of innovation in the field of resource exploration. At present, generative AI technology in the field of geoscience applications still exists in data completeness defects, geological complex system challenges, geological model interpretability problems and other core constraints. Comprehensive analysis points out that generative AI reconstructs the technical system of “data perception-knowledge refinement-decision generation”, which will certainly accelerate the realization of breakthroughs in the application of this technology in the fields of Earth science, and provide important technical support for the innovation of exploration technology and methodology, improvement of exploration efficiency, and assisting the new round of strategic action of finding mineral breakthroughs to safeguard the country’s energy security.

Three-dimensional prediction of concealed ore bodies is gradually becoming a key technology and method for mineral resources exploration in the deep earth. Accurately calibrating the association between ore-controlling factors and mineralization is crucial to the performance of prediction model. The formation of mineralization is a typical spatial non-stationary process. Based on exploration data, leveraging big data technology to quantitatively investigate the spatial non-stationarity in the relationship between ore-controlling factors and mineralization can not only provide more accurate key parameters for prediction model, but help clarify the genetic mechanism behind mineralization. In this paper, we take the Sanshandao gold deposit as an example to study the characteristics of spatial non-stationary influence of ore-controlling factors on mineralization. First, the 3D Geographical Weighted Regression (GWR) model is introduced to explore the spatial non-stationary influence. Second, the anisotropy is analyzed by introducing the directional factor to the weight calculation of 3D GWR. Third, the multi-scale characteristic is explored by applying the 3D multiscale GWR model (MGWR). Furthermore, the stationary index is calculated to analyze the stationarity degree of different ore-controlling factors’ influence. Subsequently, a comparative analysis of the intensity and variability of different ore-controlling factors’ influence is carried out. Finally, the interrelationship among direction, scale and intensity of each ore-controlling factor’s influence on the mineralization is further explored based on the mineralization laws.

This study conducted a GeoHealth survey in Shangshu Township, Anji County-a representative high geochemical background region characterized by the distribution of the black rock series in northwestern Zhejiang Province-with a focused investigation of the geochemical behavior of cadmium (Cd) and its potential impact on local population health. Results indicate that, according to the Soil Environmental Quality-Risk Control Standards for Soil Pollution in Agricultural Land (GB 15618-2018), cadmium concentrations in over 30% of the surface soils significantly exceed the regulatory thresholds. Elevated levels are primarily concentrated in the central and northern regions of the township, exhibiting strong spatial concordance with the distribution of the black rock series. Vertical profiling reveals a distinct surface accumulation of cadmium, suggesting that the geochemical conditions of the upper soil horizon promote Cd enrichment, thereby increasing the likelihood of regulatory exceedance. Sequential extraction analysis indicates that cadmium in agricultural soils exists predominantly in the ion-exchangeable form, followed by humic acid-bound, residual, iron-manganese oxide-bound, carbonate-bound, organic-bound, and water-soluble forms. The dominance of the ion-exchangeable fraction indicates high bioavailability and mobility, implying a substantial potential for environmental migration and biological uptake. Health risk assessments suggest an elevated carcinogenic risk. However, regional epidemiological data-including life expectancy, chronic disease prevalence, and skeletal health indicators, which are comparable to those in low-cadmium background areas, show no significant adverse health outcomes that can be attributed to the naturally elevated cadmium background. These findings highlight the limitations of inferring human health risks solely based on geogenic cadmium concentrations. As cadmium uptake and toxicological effects are influenced by multiple environmental and physiological factors, health risk evaluations in areas with high natural geochemical backgrounds should rely primarily on empirical regional epidemiological health data rather than geochemical levels alone.

The northeastern margin of the Qinghai-Tibet Plateau, characterized by intricate geological architecture and concentrated agro-pastoral activities, presents an optimal field laboratory for health geology research. This study systematically investigated the spatial distribution, enrichment mechanisms, and health potential of critical essential elements (Zn, Se, Ge) using advanced geochemical prospecting techniques, establishing a scientific foundation for optimizing land resource utilization and delineating priority eco-health zones. A total of 8273 surface soil samples were collected and analyzed for pH, Zn, Se, and Ge content. Multi-criteria assessment frameworks incorporating the geo-accumulation index (I_geo), standard threshold comparison, health potential index (HPI), and a Composite Index were implemented. Key findings include: (1) Soils in the study area exhibit overall low Se, low Ge, and high Zn backgrounds, with spatial variability characterized by strong variation for Se, moderate for Ge, and weak for Zn. (2) Relative to Chinese soil background values, the degree and spatial extent of Zn, Se, and Ge enrichment decrease in that order. Compared to relevant national standards and soil classification criteria, soils are generally Zn-rich but Se-poor and Ge-poor, although localized Se-rich and Ge-rich zones exist, exhibiting potential health benefits. Health potential assessment (HPI) indicates moderate potential for Zn and Se, but lower potential for Ge. Areas with moderate or higher composite health potential constitute 14.8% of the region. (3) Priority target areas for utilization were identified by first selecting zones exceeding the 85th percentile of the Composite Health Potential Index, followed by refinement based on geological background, land use, geomorphology, and enrichment patterns. Five key target areas were delineated, providing strategic direction for in-depth healthy geology research and application in Qinghai.

Certain phenolic compounds, such as bisphenol A and alkylphenols, are recognized endocrine-disrupting chemicals (EDCs) and emerging contaminants of global concern. They exhibit teratogenic, carcinogenic, and mutagenic properties, posing significant threats to ecosystems and human health. This study analyzed ten phenolic EDCs in 33 water samples (surface water, groundwater, leachate) from three North China cities using liquid chromatography-triple quadrupole-linear ion trap mass spectrometry (LC-QqQ-LIT-MS). Detection frequencies were 72% (surface water), 35% (groundwater), and 89% (leachate). Bisphenol A (BPA) reached maximum concentrations of 622ng/L (surface water), 21.8ng/L (groundwater), and 753ng/L (leachate). Ecological risk assessment for specific EDCs via the U.S. EPA risk quotient (RQ) method indicated high risk in surface water but low risk in groundwater. Pollutant profiles showed significant correlations between surface water and groundwater across cities, with consistent concentration trends suggesting potential contamination via surface water infiltration. Continuous monitoring of phenolic EDCs in groundwater is essential to assess their environmental persistence and health impacts.

Dongshan and Xishan in Suzhou are China’s primary loquat production regions. Investigating the distribution and differentiation of mineral elements in loquat fruits from these areas and evaluating health risks associated with loquat consumption are crucial for improving fruit quality and ensuring food safety. Orthogonal partial least squares discriminant analysis (OPLS-DA) revealed significant differences in elemental composition between fruits from the two regions. Selenium (Se), cobalt (Co), and manganese (Mn) contents were higher in Xishan fruits, whereas magnesium (Mg) and copper (Cu) were higher in Dongshan fruits. Partial least squares regression (PLSR) indicated a close relationship between fruit and local soil elemental contents, with distinct correlation patterns between regions: In Xishan: Fruit Zn showed positive correlations with soil Mg, Mn, and Ni, but a negative correlation with soil I; Fruit Mn was positively correlated with soil Mn and Cu, and negatively correlated with soil I, Cd, and Cr. In Dongshan: Fruit Zn showed positive correlations with soil Ca, Mn, and As, but negative correlations with soil Cu and Cd; Fruit Mn was positively correlated with soil Ni, Cd, and As, and negatively correlated with soil Zn and Cr. The two regions lie within the Guanshan Formation (Xishan) and Maoshan Formation (Dongshan), respectively. The distinct parent materials of these geological formations are likely the fundamental cause of the observed elemental differences in loquat fruits. Heavy metal contents in fruits from both regions complied with GB 2762—2022 limits, with both the target hazard quotient (THQ) and potential carcinogenic risk index (CRI) well below safety thresholds. Local soils met the environmental quality standards for green food production areas, supporting loquat cultivation for green food. These findings provide a scientific basis for loquat cultivation planning and elemental regulation in Suzhou.

In this study, we developed an integrated pollution source apportionment framework to investigate nitrate origins in groundwater within the Liaodong Bay region. By analyzing 51 shallow groundwater samples, we employed the Expanded Durov diagram, Pearson correlation analysis, and random forest algorithm to identify nine monitoring indicators significantly correlated with nitrate concentrations, including TDS, {\mathrm{SO}}_4^{2-}, and Ca²⁺. The Positive Matrix Factorization (PMF) method was applied to quantify nitrate pollution sources in typical mountainous areas. Results revealed that the predominant hydrochemical type is HCO₃·Cl-Na·Ca, with a nitrate exceedance rate of 20% (relative to GB/T 14848-2017) and pronounced spatial heterogeneity. Source apportionment of nitrate contamination identified two dominant sources: agricultural fertilizer application (57.27%) and processes associated with seawater intrusion (42.52%). Key indicators such as TDS, {\mathrm{SO}}_4^{2-}, and Ca²⁺ exhibit strong co-variation patterns, indicating shared pollution pathways. The integrated framework of “multi-method integration, screening of correlated indicators, and quantitative source apportionment” established in this study provides a scientific foundation for optimizing groundwater pollution monitoring networks and formulating targeted mitigation strategies for nitrate in coastal regions.

We systematically study the characteristics of zircon U-Pb geochronology, whole-rock major and trace elements, in-situ apatite trace elements and Nd isotopes, and in-situ zircon Hf isotopes of the Jiguanshan granite porphyry in the Liaodong Peninsula, northeastern North China Craton, aiming to reevaluate its emplacement age and petrogensis, and further to constrain its tectonic significance. The zircon grains in the Jiguanshan granite porphyry show euhedral to subhedral textures, typical oscillatory growth zoning, and high Th/U(0.27-1.12) ratios, indicating a magmatic origin. The zircon U-Pb dating results of fifteen magmatic zircon grains illustrate that the Jiguanshan granite porphyry formed in Early Cretaceous(126±1.4 Ma). The Jiguanshan granite porphyry shows relatively high SiO₂(75.70%-76.07%) and alkali(Na₂O+K₂O, 7.88%-8.30%) but low MgO(0.08%-0.15%) contents (mass fraction). The aluminum saturation index of the Jiguanshan granite porphyry ranges from 1.19 to 1.30, with a peraluminous characteristics. In addition, the Jiguanshan granite porphyry is enriched in light rare earth elements(LREEs), but depleted in heavy rare earth elements(HREEs) with distinctly negative Eu anomalies(Eu/Eu^*=0.36-0.51). Furthermore, the Jiguanshan granite porphyry exhibits a typical arc affinity, with enrichment in large ion lithophile elements(LILEs; e.g., Rb, Th, U) and depletion in high field strength elements(HFSEs; e.g., Nb, Ta). The apatite grains in the Jiguanshan granite porphyry are of magmatic origin. They are enriched in LREEs and LILEs, and depleted in HREEs and HFSEs, with distinctly negative Eu anomalies(Eu/Eu^*=0.11-0.24), which are similar with its host rock. The Jiguanshan granite porphyry belongs to highly fractionated I-type granite. Combining with the relatively low zircon ε_Hf(t) values(-15.5 to -20.1), and apatite ε_Nd(t) values(-15.0 to -16.7), as well as relatively old two-stage Nd-Hf model age (2454-2135 Ma), we believe that the Jiguanshan granite porphyry originated from partial melting of the Paleoproterozoic lower crust that mainly consists of meta-igneous rocks of the Liaohe Group. In combination with the extensive distribution of A-type granites, metamorphic core complexes and large-scale extensional basins in the study area, as well as the migration of magmatic activity from inland to trench during the Late Jurassic to Early Cretaceous, we conclude that the Jiguanshan granite porphyry formed in the extensional tectonic setting related to the roll-back of the subducted Paleo-Pacific plate towards the northeastern North China Craton.

The Jiaodong Peninsula, despite covering less than 10000 km², has proven gold reserves exceeding 6000 tons, and its massive gold mineralization characteristics have attracted widespread attention and research from both domestic and international scholars. Numerous dating studies show that the gold mineralization in Jiaodong is concentrated around 120±5 Ma. However, with the reporting of more varying gold mineralization age data, discussions about multi-stage mineralization have arisen, although these discussions lack evidence from geological bodies directly related to gold mineralization. This study discovered potential gold-related rocks at the Yinggezhuang gold deposit, including albite granite containing nodular and dendritic quartz and diorite porphyry coexisting with quartz-sulfide veins. Through petrographic analysis, precise gold, silver, sulfur content determination, and dating studies, the albitite granite was characterized by high sulfur ((17-2134)×10^-6), gold (average 5.2×10^-9), and volatile content. These features suggest that at the late stage of high-fractionation granite magma evolution, molten bodies rich in silica, sulfur, and gold might have resulted in gold-bearing quartz-sulfide veins. The U-Pb age of the magmatic titanite in the albite granite is 144.5±1.8 Ma, while the U-Pb age of zircons from the diorite porphyry that nearly simultaneously formed with the quartz-sulfide vein is 147.1±0.58 Ma, indicating that the Jiaodong region may have experienced gold mineralization events related to alkaline magmatic activity around 147-144 Ma.

The Yunkai low uplift is located between two hydrocarbon-rich depressions, with good prospects for oil and gas exploration. It plays a critical role in understanding the differential evolution of the eastern and western Pearl River Mouth Basin. Based on 3D seismic data in the depth domain, this paper focused on the geological structures, fault characteristics and tectonic evolution process of the Yunkai low uplift by combining the characteristics of the growth strata, the results of tectonic evolution profiles and simulations of the subsidence history, as well as analysing the dynamic environment of the tectonic evolution and tectonic zoning role of low uplift. The Yunkai low uplift can be divided into three sections from north to south, with different structural patterns in each section, and the contacts between different sections and between them and depressions are mostly faults. The results of the growth strata characteristics analysis and the simulation of the uplift-descent response features of the low uplift and the depressions on both sides reveal that there are spatial and temporal variations in the uplift rate at different sections of the Yunkai low uplift. The geometric and kinematic characteristics of the faults in the Cenozoic rifting and post-rifting stages are different, and the fault strikes undergo clockwise rotation from early to late. Two-stage late Mesozoic fault systems with extrusion or compression-torsion properties can be identified within the basement and have significantly constrained the Cenozoic faults. Overall, the low uplift experienced two stages of extrusion deformation in the Late Jurassic-Early Cretaceous NW and early-middle Late Cretaceous near-SN directions during the late Mesozoic. It experienced three major formation stages in the Cenozoic: a rapid uplift stage during the Eocene, a slow uplift stage during the Late Eocene-Early Miocene, and a whole sedimentation-subsidence and deep burial stage from the Miocene to the present. Comprehensive analyses suggest that the Yunkai Low uplift, as a tectonic transition zone between the depressions on both sides, is deeply and shallowly superimposed on the southern section of the NW-trending Yangjiang-Yitong Ansha fault zone, which regulates the differential structural deformation of the depressions and the differential evolution of the basin. The NW-trending structural transition zone in which the Yunkai low uplift is situated has a significant deep dynamic setting, and this property results in the partitioning of the dominant Cenozoic NE-trending structures in the basin.

The Anninghe fault zone is a seismically active fault zone in western Sichuan characterized by strong earthquake activity. To investigate the lithology, physical properties, fine fault structure, and spatiotemporal variations of the fault medium, we applied cross-well seismic tomography to image the fine structure of the fault zone. The results derived using the bent-ray tracing method reveal a P-wave velocity structure image of the Anninghe fault zone that aligns well with fault features identified from borehole drilling and logging data. The acquired high-resolution velocity structural image provides crucial constraints for understanding the seismogenic processes of earthquakes on the Anninghe fault zone and for assessing its potential strong earthquake hazards.

Strike-slip faults are well developed in the northern depression of the Tarim Basin, playing a crucial role in controlling the development of fault-controlled reservoirs in the Ordovician carbonate rocks in the eastern Aman Transition Zone. Based on 3D seismic data and Ordovician geological information, and from the perspective of reservoir characterization and development needs of the No.19 strike-slip fault-controlled oil reservoir, this study conducts a detailed analysis of the layered deformation characteristics, segmented deformation features, activity patterns, evolutionary stages, and fault evolution process of the No.19 strike-slip fault zone, with particular emphasis on its reservoir-controlling significance. The main findings are as follows: (1) The No.19 strike-slip fault zone can be divided into four structural layers from bottom to top: the subsalt Cambrian structural layer, Middle Cambrian evaporite structural layer, Middle-Lower Ordovician carbonate structural layer, and Silurian clastic structural layer. (2) Based on fault strike and combination characteristics of faults at the top of Ordovician carbonate rocks, the fault zone is classified into horsetail, en echelon, braided, overlapping, and linear segments. (3) The Middle-Lower Ordovician section of the No.19 strike-slip fault zone exhibits strong overall activity with significant deformation, and can be subdivided into seven transtensional segments, seven transpressional segments, and three translational segments. The fault zone underwent multistage tectonic evolution, with the Early and Middle Caledonian being the key active periods. (4) Different planar segmentation patterns exert varying controls on reservoir development: braided and overlapping segments show relatively better reservoir development, while horsetail, en echelon, and linear segments exhibit poorer reservoir development. Profile structural styles significantly influence reservoir development, with transtensional segments favoring large-scale reservoir formation, transpressional segments promoting reservoir development, and translational segments yielding limited reservoir scale. The superposition of planar segmentation patterns and profile structural styles leads to notable differences in hydrocarbon accumulation scale across different superimposed segments of the same fault zone. (5) The layered deformation, planar segmentation patterns, activity characteristics, and profile structural styles collectively contribute to variations in reservoir development along different sections of the same fault zone. Research significance: This study provides an important structural foundation for investigating the vertical development patterns and connectivity of strike-slip fault-controlled reservoirs, and offers valuable insights for further research on the reservoir-controlling significance of strike-slip fault-controlled reservoirs.

The exploration and development of deep shale gas is crucial for achieving the dual-carbon target. To achieve this, multi-clustered hydraulic fracturing is an important method for enhancing production on a large scale. However, when the perforations are densely laid, the propagation of some clusters may be limited. Therefore, it is important to determine a reasonable spacing of fractures to improve reserve recovery and reduce costs. In this study, we established a hydraulic fracturing model for multi-layered reservoirs considering natural fractures. We used the displacement discontinuity method and the P3D model, and introduced Broyden iterative calculation method to propose an efficient solution method for the fluid-solid coupled multi-fracture model. We simulated and analyzed the effects of horizontal stress difference and construction parameters on the morphology of multi-cluster fractures and the uniformity of fracture propagation. The results showed that Broyden iteration is more computationally efficient than Newton iteration. Hydraulic fractures in deep shale reservoirs are difficult to turn under high stress difference conditions (10 MPa). Increasing the cluster spacing to 8m significantly improves the uniformity of each cluster. Additionally, raising the viscosity (>20 mPa·s) and pumping rate (>16 m³/min) of the fracturing fluid is beneficial for dense and uniform fracture propagation. These research results provide theoretical support for optimizing the deep shale fracturing process in China.

Due to multi-phase karstification superimposed over different geological periods, the Ordovician in the Tahe area of the Tarim Basin has developed paleokarst conduit systems characterized by diverse types, complex morphologies, and intricate structures, demonstrating significant exploration and development potential. By integrating 3D seismic data, well logging data, seismic attribute extraction, and wave impedance inversion technologies, this study reconstructs the paleo-tectonic, paleo-geomorphic, and paleo-hydrological conditions of the Ordovician in the Tahe Oilfield. The planar morphology and vertical stratification of large-scale karst conduits are systematically analyzed, and their unique genetic mechanisms are elucidated by identifying the key controlling factors governing conduit development. Based on these findings, a comprehensive geological model of karst conduits is established. The results reveal that the Ordovician formation in the Tahe Oilfield hosts 13 large-scale epigenic karst conduits systems, exhibiting diverse planar configurations such as single-branch, dendritic, and maze-like patterns. Vertically, seven correlative cave layers are identified, with development depth progressively increasing from east to west. The structural heterogeneity of these karst conduits is closely associated with variations in hydrodynamic field types. Four distinct geological models of conduit development are proposed for the Ordovician in the Tahe Oilfield: “Anticlinal confluence type” “Karst canyon type” “Single-bevel flow type” and “Syncline flow type”. The proposed models offer new perspectives on Ordovician karst conduit evolution in the Tahe Oilfield, serving as both a theoretical benchmark and practical tool for hydrocarbon exploitation.

The Early Cretaceous was marked by extensive wildfires across the globe. Wildfires, as an critical component of the global ecosystem, are increasingly recognized for their profound impact on deep-time ecosystems. Based on the wildfire records from the Early Cretaceous in NE China, as well as the global records on CO₂, O₂ concentrations and the evolution of vegetation, this study synthesizes the characteristics and influencing factors of wildfires and analyzes their implications on deep-time ecosystems during this period. The Early Cretaceous Albian in NE China exhibits an inertinite content of 24.17% (volume fraction), compared to a lower content of 18.68% (volume fraction) in the Aptian. Based on the relationship model between inertinite content and oxygen content, the estimated Albian oxygen concentration is 24.92% and Aptian oxygen concentration is 24.21%. Utilizing a relationship model that correlates inertinite reflectance with combustion temperature, the types of wildfires in the Early Cretaceous Albian and Aptian stages in NE China are classified as ground fires and surface fires. From the Aptian to the Albian stages, the types of wildfires are predominantly ground fires. During the early phase of the Early Cretaceous, angiosperms were characterized by their low stature, shorter life cycles, high vein density, and high photosynthetic rates, which better adapted them to the low-intensity wildfires during this period. Meanwhile, the frequent occurrence of wildfires in the Early Cretaceous promoted the wide spread of angiosperms, which in turn led to a reduction in the diversity of gymnosperms and ferns. Wildfires led to enhanced surface erosion and runoff, which facilitated the influx of large amounts of terrestrial organic matter and nutrients, such as nitrogen and phosphorus, into lakes or oceans. This influx caused eutrophication in lakes or oceans and promoted the proliferation of planktonic organisms. As substantial amounts of terrestrial organic matter and plankton sink in the water, they consumed dissolved oxygen, fostering the development of oxygen-deficient environments. This process contributed to the accumulation of organic-rich mudstones during the Early Cretaceous.

In this paper, the shallow groundwater in the North China Plain is taken as the object of study, and based on the zoning of the groundwater system and the history of sea intrusion, as well as the characteristics of the iodine concentration distribution, four typical zones are delineated:Haihe Plain Groundwater System Zone (A); Haihe Plain Groundwater System-Sea Intrusion Zone (B); Lower Paleo-Yellow River Groundwater System Zone (C); and Lower Paleo-Yellow River Groundwater System-Sea Intrusion Zone (D). Overall, the shallow groundwater in the North China Plain was nearly neutral to alkaline,and the distribution of iodine ion concentration was clearly zoned, with a clear upward trend in iodine ion concentration from the pre-mountainous area to the coastal area, and the hydrochemical type of the water also appeared to vary from HCO₃-Ca type water with a low TDS (A) to HCO₃-Na type water with a TDS of more than 1 g/L (C),to Cl-Na type water with a TDS of more than 3 g/L (B,D). The high iodine groundwater were mainly distributed in zones B, C, and D, with mean iodine ion concentrations of 128.27 μg/L, 176.7 μg/L, and 179.2 μg/L respectively, and the median values of zones C and D were more than 100 μg/L. For the three high iodine zones, we screened the iodine-influencing factors and preliminarily explored the causes of the high iodine groundwater in the study area by using the method of random forest. The results showed that Zone C belongs to the medium TDS-high iodine zone, where the complex clay and sand interlayer phenomenon was formed due to the deposition of the Yellow River in multiple flooding periods, the groundwater circulation is weak, the concentration of iodine ions is related to the content of organic matter, clay minerals and iron oxides in the alluvial lake sediments, and the main role of the formation of high iodine groundwater is the evaporation and concentration. Zone D belongs to the high TDS-high iodine zone, and the sea invasion events in the past made a large amount of iodine stored in the sediments, and the formation of high iodine groundwater is mainly due to evaporation and concentration. Area D is a high TDS-moderate iodine area, and the historical marine intrusion events have caused a large amount of iodine to be stored in the sediments, providing a source of iodine in the groundwater, and the biodegradation of organic matter is the main driving factor for the formation of high iodine groundwater in this area. Area B belongs to the high TDS-moderate iodine area, and the marine sediment formed by the historical marine intrusion events is also the source of iodine in the aquifer in this area, and the iodine concentration of the area is slightly lower than that of Area D because of leaching from the lower tip of the Nine Rivers.

The identification of groundwater pollution risk under ecological replenishment conditions is of great significance for groundwater safety replenishment. Although extensive research has been carried out on groundwater pollution risk at present, the evolution of groundwater pollution risk under ecological replenishment conditions is rarely involved. Therefore, this paper takes the Yongding River recharge area as the research object, starts with the dynamic changes of industrial pollution source pattern and groundwater level before and after the recharge(2002-2022), and constructs a groundwater pollution risk assessment system with coupled pollution load input-vadose zone restrain-aquifer transport, discusses the evolution law of groundwater pollution risk driven by pollution load and water level fluctuation, and provides support for the prevention and control of groundwater pollution risk. The results show that the overall number and load of industrial pollution sources show a downward trend from 2002 to 2022, and the main reason is the closure and relocation of smelting and processing industry represented by Shougang Industrial Park. By 2022, the high and medium risk load areas decrease to 0 and 14% respectively. In contrast, groundwater vulnerability showed a trend of decreasing first and then increasing. Before 2014, overmining in successive years led to a continuous decline in water level. Subsequently, reduced mining initiated at the end of 2014 and external water ecological restoration starting in 2019 led to a rapid rise in water level, and the dynamic change of water level directly affected groundwater vulnerability. The comprehensive risk of groundwater pollution in the study area showed an evolution law of decreasing first and then increasing, and 2014 was the turning point. Prior to 2014, the risk was mainly affected by the dual factors of surface pollution load and water level fluctuation. The change of industrial pattern led to the reduction of pollution load, and the decrease of water level led to the reduction of groundwater vulnerability, so the comprehensive risk continued to decline. With the commencement of reduced mining at the end of 2014 and the external water ecological restoration in 2019, the rapid rise of water level led to increased groundwater vulnerability. However, as the pattern of surface pollution load did not change significantly, the comprehensive risk increased, and water level fluctuation became the main factor at this stage.Enterprises with medium risk or above were mainly distributed in Hedong and Fengtai. The proportion of such enterprises classified as medium or high risk increased from 3% in 2014 to 15% in 2022.