Computers & Geosciences最新文献_第6页

Uncertainty-aware ensemble learning and dynamic threshold optimization for landslide susceptibility mapping 滑坡易感性制图的不确定性感知集成学习与动态阈值优化

IF 4.4 2区地球科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Geosciences

Pub Date : 2025-08-29 DOI: 10.1016/j.cageo.2025.106042

Ting Xiao , Wei Huang , Lichang Wang , Beibei Yang , Zuohui Qin , Xiaodong Liu , Yingbin Xiao

Landslides represent a prevalent and devastating geological hazard. Identifying areas susceptible to landslides is vital for disaster prevention and reduction. However, traditional models suffer from limited predictive accuracy, strong regularity in breakpoint selection for susceptibility zoning, and inconsistent predictions across different models, resulting in uncertainty in susceptibility assessment. To address these issues, this study proposes an innovative intelligent landslide susceptibility mapping approach that integrates ensemble learning, multi-model uncertainty analysis, and dynamic optimization. Focusing on Linxiang City, Hunan Province, China, this research synthesizes historical landslide inventories and field-identified unstable slopes as positive samples. Three base models were constructed: logistic regression (LR), random forest (RF), and graph neural network (GNN). Ensemble learning using the stacking method was applied to combine these models. The ensemble further incorporates prediction uncertainty estimation and multi-dimensional k-nearest neighbor (KNN) adjacency matrix. Utilizing an attention mechanism, the model dynamically integrates geographic features, environmental factors, and prediction outputs. The final output is a prediction model that synthesizes spatial structure information and prediction uncertainties. For susceptibility mapping, this study proposes a dynamic optimization approach combining Natural Breaks, Frequency Ratio, and Equal Interval methods, determining optimal threshold combinations through relative density distribution of landslide occurrences to enhance susceptibility classification rationality. Model performance was evaluated and compared using area under roc curve (AUC), where a larger AUC signifies higher predictive accuracy. The results show that the ensemble model outperformed all others with an AUC of 0.95, compared to the base models' AUCs of 0.82 (LR), 0.84 (RF), and 0.87 (GNN). This demonstrates that the ensemble learning methods that incorporate uncertainty achieve higher accuracy in risk identification than conventional models. The dynamic classification method also shows a better performance over conventional approaches in high-susceptibility classification precision and landslide density differentiation.

滑坡是一种普遍存在的破坏性地质灾害。确定易受滑坡影响的地区对防灾减灾至关重要。然而，传统模型的预测精度有限，易感性分区断点选择的规律性强，不同模型之间的预测结果不一致，导致易感性评估存在不确定性。为了解决这些问题，本研究提出了一种集成集成学习、多模型不确定性分析和动态优化的滑坡敏感性智能制图方法。本研究以湖南省临乡市为研究对象，综合了历史滑坡清单和现场鉴定的不稳定边坡为阳性样本。构建了逻辑回归（LR）、随机森林（RF）和图神经网络（GNN）三种基本模型。采用集成学习的叠加方法对这些模型进行组合。该集成进一步结合了预测不确定性估计和多维k近邻（KNN）邻接矩阵。利用注意机制，该模型动态集成了地理特征、环境因素和预测输出。最后的输出是一个综合了空间结构信息和预测不确定性的预测模型。对于敏感性映射，本文提出了一种结合自然断裂法、频率比法和等间隔法的动态优化方法，通过滑坡发生点的相对密度分布确定最优阈值组合，以提高敏感性分类的合理性。使用roc曲线下面积（AUC）评估和比较模型性能，其中AUC越大表示预测精度越高。结果表明，与基础模型的AUC分别为0.82 （LR）、0.84 （RF）和0.87 （GNN）相比，集成模型的AUC为0.95，优于其他所有模型。这表明集成学习方法在风险识别方面比传统模型具有更高的准确性。动态分类方法在高敏感性分类精度和滑坡密度分异方面也优于常规方法。

{"title":"Uncertainty-aware ensemble learning and dynamic threshold optimization for landslide susceptibility mapping","authors":"Ting Xiao , Wei Huang , Lichang Wang , Beibei Yang , Zuohui Qin , Xiaodong Liu , Yingbin Xiao","doi":"10.1016/j.cageo.2025.106042","DOIUrl":"10.1016/j.cageo.2025.106042","url":null,"abstract":"<div><div>Landslides represent a prevalent and devastating geological hazard. Identifying areas susceptible to landslides is vital for disaster prevention and reduction. However, traditional models suffer from limited predictive accuracy, strong regularity in breakpoint selection for susceptibility zoning, and inconsistent predictions across different models, resulting in uncertainty in susceptibility assessment. To address these issues, this study proposes an innovative intelligent landslide susceptibility mapping approach that integrates ensemble learning, multi-model uncertainty analysis, and dynamic optimization. Focusing on Linxiang City, Hunan Province, China, this research synthesizes historical landslide inventories and field-identified unstable slopes as positive samples. Three base models were constructed: logistic regression (LR), random forest (RF), and graph neural network (GNN). Ensemble learning using the stacking method was applied to combine these models. The ensemble further incorporates prediction uncertainty estimation and multi-dimensional k-nearest neighbor (KNN) adjacency matrix. Utilizing an attention mechanism, the model dynamically integrates geographic features, environmental factors, and prediction outputs. The final output is a prediction model that synthesizes spatial structure information and prediction uncertainties. For susceptibility mapping, this study proposes a dynamic optimization approach combining Natural Breaks, Frequency Ratio, and Equal Interval methods, determining optimal threshold combinations through relative density distribution of landslide occurrences to enhance susceptibility classification rationality. Model performance was evaluated and compared using area under roc curve (AUC), where a larger AUC signifies higher predictive accuracy. The results show that the ensemble model outperformed all others with an AUC of 0.95, compared to the base models' AUCs of 0.82 (LR), 0.84 (RF), and 0.87 (GNN). This demonstrates that the ensemble learning methods that incorporate uncertainty achieve higher accuracy in risk identification than conventional models. The dynamic classification method also shows a better performance over conventional approaches in high-susceptibility classification precision and landslide density differentiation.</div></div>","PeriodicalId":55221,"journal":{"name":"Computers & Geosciences","volume":"206 ","pages":"Article 106042"},"PeriodicalIF":4.4,"publicationDate":"2025-08-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144925125","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Controlled latent diffusion models for 3D porous media reconstruction 三维多孔介质重建的可控潜扩散模型

IF 4.4 2区地球科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Geosciences

Pub Date : 2025-08-26 DOI: 10.1016/j.cageo.2025.106038

Danilo Naiff , Bernardo P. Schaeffer , Gustavo Pires , Dragan Stojkovic , Thomas Rapstine , Fabio Ramos

Three-dimensional digital reconstruction of porous media presents a fundamental challenge in geosciences, requiring simultaneous resolution of fine-scale pore structures while capturing representative elementary volumes. This work introduces a computational framework that addresses this challenge through latent diffusion models operating within the Elucidated Diffusion Models (EDM) framework. The proposed approach reduces dimensionality via a custom Variational Autoencoder trained in binary geological volumes, improving efficiency and also enabling the generation of larger volumes than previously possible with diffusion models. A key innovation is the controlled unconditional sampling methodology, which enhances distribution coverage by first sampling target statistics from their empirical distributions, and then generating samples conditioned on these values. Extensive testing on four distinct rock types demonstrates that conditioning on porosity – a readily computable statistic – is sufficient to ensure a consistent representation of multiple complex properties, including permeability, two-point correlation functions, and pore size distributions. The framework achieves better generation quality than pixel-space diffusion while enabling significantly larger volume reconstruction (

25 6^{3}

voxels) with substantially reduced computational requirements, establishing a new state-of-the-art for digital rock physics applications.

多孔介质的三维数字重建是地球科学领域的一个基本挑战，它需要在捕获代表性基本体积的同时，对细尺度孔隙结构进行分辨率处理。这项工作引入了一个计算框架，通过在阐明扩散模型（EDM）框架内运行的潜在扩散模型来解决这一挑战。该方法通过自定义的变分自编码器（Variational Autoencoder）减少了二进制地质体积的维数，提高了效率，并且能够生成比以前使用扩散模型更大的体积。一个关键的创新是受控无条件抽样方法，它通过首先从其经验分布中抽样目标统计量，然后根据这些值生成样本来增强分布覆盖率。对四种不同岩石类型的广泛测试表明，孔隙度（一个易于计算的统计数据）的调节足以确保多种复杂属性的一致表示，包括渗透率、两点相关函数和孔隙大小分布。该框架实现了比像素空间扩散更好的生成质量，同时实现了更大的体积重建（2563体素），大大减少了计算需求，为数字岩石物理应用建立了新的技术水平。

{"title":"Controlled latent diffusion models for 3D porous media reconstruction","authors":"Danilo Naiff , Bernardo P. Schaeffer , Gustavo Pires , Dragan Stojkovic , Thomas Rapstine , Fabio Ramos","doi":"10.1016/j.cageo.2025.106038","DOIUrl":"10.1016/j.cageo.2025.106038","url":null,"abstract":"<div><div>Three-dimensional digital reconstruction of porous media presents a fundamental challenge in geosciences, requiring simultaneous resolution of fine-scale pore structures while capturing representative elementary volumes. This work introduces a computational framework that addresses this challenge through latent diffusion models operating within the Elucidated Diffusion Models (EDM) framework. The proposed approach reduces dimensionality via a custom Variational Autoencoder trained in binary geological volumes, improving efficiency and also enabling the generation of larger volumes than previously possible with diffusion models. A key innovation is the controlled unconditional sampling methodology, which enhances distribution coverage by first sampling target statistics from their empirical distributions, and then generating samples conditioned on these values. Extensive testing on four distinct rock types demonstrates that conditioning on porosity – a readily computable statistic – is sufficient to ensure a consistent representation of multiple complex properties, including permeability, two-point correlation functions, and pore size distributions. The framework achieves better generation quality than pixel-space diffusion while enabling significantly larger volume reconstruction (<span><math><mrow><mn>25</mn><msup><mrow><mn>6</mn></mrow><mrow><mn>3</mn></mrow></msup></mrow></math></span> voxels) with substantially reduced computational requirements, establishing a new state-of-the-art for digital rock physics applications.</div></div>","PeriodicalId":55221,"journal":{"name":"Computers & Geosciences","volume":"206 ","pages":"Article 106038"},"PeriodicalIF":4.4,"publicationDate":"2025-08-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144907447","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Segmentation of stochastic scalar fields in unstructured meshes 非结构化网格中随机标量场的分割

IF 4.4 2区地球科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Geosciences

Pub Date : 2025-08-26 DOI: 10.1016/j.cageo.2025.106041

Tommaso Sorgente , Marianna Miola , Simone Pittaluga , Daniela Cabiddu , Michela Mortara , Marino Vetuschi Zuccolini

We present an algorithm for segmenting a (stochastic) scalar field defined on an unstructured mesh into a given number of parts. It can be applied to any type of mesh, such as triangular/tetrahedral meshes, 2D/3D grids, and generic polygonal/polyhedral meshes, inducing a classification of the mesh elements into regions with limited noise and smooth boundaries. The algorithm offers multiple output options, providing valuable information about the segmentation and the mesh regions in various file formats, thus making it suitable for practical applications. We show the algorithm at work in different application scenarios, ranging from environmental geochemistry to marine sciences and groundwater modeling, proving its efficacy and versatility.

我们提出了一种将定义在非结构化网格上的（随机）标量场分割为给定数量的部分的算法。它可以应用于任何类型的网格，如三角形/四面体网格、2D/3D网格和一般多边形/多面体网格，将网格元素分类到具有有限噪声和光滑边界的区域。该算法提供了多种输出选项，以各种文件格式提供了关于分割和网格区域的有价值的信息，因此适合实际应用。我们展示了该算法在不同应用场景中的工作，从环境地球化学到海洋科学和地下水建模，证明了它的有效性和多功能性。

引用次数: 0

SmartMagDL: Smartphone geomagnetic mapping using deep learning SmartMagDL：使用深度学习的智能手机地磁制图

IF 4.4 2区地球科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Geosciences

Pub Date : 2025-08-23 DOI: 10.1016/j.cageo.2025.106040

Elad Fisher , Roger Alimi , Miki Vizel , Itzik Klein

Magnetic field mapping is an essential tool in geoscience, for identifying anomalies and understanding subsurface structures, requiring systematic and methodical data acquisition. The use of smartphones’ built-in magnetometers for this task offers advantages such as cost-effectiveness, accessibility, and simplicity. Recent works relied on model-based interpolation techniques significantly limited by sparse data collection, sensor noise, orientation-dependent distortions, and overall low data quality. As a result, magnetic maps were often noisy and unreliable for practical applications. In this work, we aim to fill this gap by introducing a deep learning (DL) approach to overcome these challenges and produce accurate, high-resolution magnetic field maps from smartphone data. To address the limitations of extensive real-world data collection, we developed an innovative two-stage simulation framework to generate the required training datasets. First, the theoretical magnetic field produced by ferromagnetic objects in a 30 m × 30 m area was computed to serve as ground truth data for the network. Second, a simulation model was implemented to replicate the data acquisition process of smartphone magnetometers. This model included real-world survey protocols, noise factors, sensor behavior, and simulated trajectories based on real world recorded data. Compared to the model-based baseline, our method improves anomaly localization, reduces noise, and enhances accuracy. At the 80th percentile the MSE and LPIPS metrics showed 75% and 55% improvements respectively, further validated by visual analysis of the reconstructed maps.

磁场测绘是地球科学中识别异常和了解地下结构的重要工具，需要系统和有条不紊的数据采集。使用智能手机内置的磁力计完成这项任务具有成本效益、可访问性和简单性等优势。最近的工作依赖于基于模型的插值技术，这些技术明显受到稀疏数据收集、传感器噪声、方向相关失真和整体低数据质量的限制。因此，磁图在实际应用中往往存在噪声和不可靠性。在这项工作中，我们的目标是通过引入深度学习（DL）方法来填补这一空白，以克服这些挑战，并从智能手机数据中生成准确、高分辨率的磁场图。为了解决广泛的现实世界数据收集的局限性，我们开发了一个创新的两阶段模拟框架来生成所需的训练数据集。首先，计算30 m × 30 m区域内铁磁物体产生的理论磁场，作为网络的地真值数据；其次，建立了智能手机磁强计数据采集过程的仿真模型。该模型包括真实世界的调查协议、噪声因素、传感器行为以及基于真实世界记录数据的模拟轨迹。与基于模型的基线相比，该方法改进了异常定位，降低了噪声，提高了精度。在第80百分位，MSE和LPIPS指标分别显示75%和55%的改善，通过重建地图的视觉分析进一步验证。

{"title":"SmartMagDL: Smartphone geomagnetic mapping using deep learning","authors":"Elad Fisher , Roger Alimi , Miki Vizel , Itzik Klein","doi":"10.1016/j.cageo.2025.106040","DOIUrl":"10.1016/j.cageo.2025.106040","url":null,"abstract":"<div><div>Magnetic field mapping is an essential tool in geoscience, for identifying anomalies and understanding subsurface structures, requiring systematic and methodical data acquisition. The use of smartphones’ built-in magnetometers for this task offers advantages such as cost-effectiveness, accessibility, and simplicity. Recent works relied on model-based interpolation techniques significantly limited by sparse data collection, sensor noise, orientation-dependent distortions, and overall low data quality. As a result, magnetic maps were often noisy and unreliable for practical applications. In this work, we aim to fill this gap by introducing a deep learning (DL) approach to overcome these challenges and produce accurate, high-resolution magnetic field maps from smartphone data. To address the limitations of extensive real-world data collection, we developed an innovative two-stage simulation framework to generate the required training datasets. First, the theoretical magnetic field produced by ferromagnetic objects in a 30 m × 30 m area was computed to serve as ground truth data for the network. Second, a simulation model was implemented to replicate the data acquisition process of smartphone magnetometers. This model included real-world survey protocols, noise factors, sensor behavior, and simulated trajectories based on real world recorded data. Compared to the model-based baseline, our method improves anomaly localization, reduces noise, and enhances accuracy. At the 80th percentile the MSE and LPIPS metrics showed 75% and 55% improvements respectively, further validated by visual analysis of the reconstructed maps.</div></div>","PeriodicalId":55221,"journal":{"name":"Computers & Geosciences","volume":"206 ","pages":"Article 106040"},"PeriodicalIF":4.4,"publicationDate":"2025-08-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144903468","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

EffiSeisM: An efficient multi-task deep learning model for earthquake monitoring effisism：用于地震监测的高效多任务深度学习模型

IF 4.4 2区地球科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Geosciences

Pub Date : 2025-08-22 DOI: 10.1016/j.cageo.2025.106039

Lixin Zhang , Ziang Li , Zhijun Dai , Hongmin Liu

Earthquake monitoring is essential for providing timely warnings, mitigating disaster impacts, advancing scientific research, and guiding urban planning. Precise seismic waveform analysis enables accurate event detection, magnitude estimation, and a deeper understanding of earthquake mechanisms. In this paper, we propose EffiSeisM, an efficient multi-task deep learning model that combines a state space model with a convolutional architecture for earthquake detection, phase picking, and magnitude estimation. EffiSeisM designed a novel Seismic Scale Conv Module and a Conv-SSM encoder, which effectively capture key seismic features while reducing computational complexity. This design ensures high accuracy and operational efficiency, enabling effective seismic analysis. We evaluate EffiSeisM on the DiTing Dataset and DiTing Dataset 2.0, comprising 3 million seismic samples from China and surrounding regions, and compare its performance with several baseline models. The results show that EffiSeisM consistently outperforms the baselines, achieving F1 scores of 0.98 for earthquake detection, 0.92 for phase-P picking, 0.84 for phase-S picking, and an R

^{2}

of 0.92 for magnitude estimation. Additionally, EffiSeisM demonstrates significant improvements in inference speed and accuracy, highlighting its potential as a scalable and efficient solution for large-scale seismic data analysis.

地震监测对于提供及时预警、减轻灾害影响、推进科学研究和指导城市规划至关重要。精确的地震波形分析使准确的事件检测、震级估计和对地震机制的更深入了解成为可能。在本文中，我们提出effisism，这是一种高效的多任务深度学习模型，它将状态空间模型与用于地震检测、相位选择和震级估计的卷积架构相结合。effisisism设计了一种新颖的地震尺度转换模块和一种convs - ssm编码器，可以有效地捕获关键地震特征，同时降低计算复杂度。这种设计确保了高精度和操作效率，实现了有效的地震分析。我们在DiTing数据集和DiTing数据集2.0上对effisism进行了评估，DiTing数据集包含来自中国及周边地区的300万地震样本，并将其性能与几个基线模型进行了比较。结果表明，effisisism始终优于基线，地震检测的F1得分为0.98，相位p拾取的F1得分为0.92，相位s拾取的F1得分为0.84，震级估计的R2为0.92。此外，effisisism在推理速度和准确性方面有了显着提高，突出了其作为大规模地震数据分析的可扩展和高效解决方案的潜力。

{"title":"EffiSeisM: An efficient multi-task deep learning model for earthquake monitoring","authors":"Lixin Zhang , Ziang Li , Zhijun Dai , Hongmin Liu","doi":"10.1016/j.cageo.2025.106039","DOIUrl":"10.1016/j.cageo.2025.106039","url":null,"abstract":"<div><div>Earthquake monitoring is essential for providing timely warnings, mitigating disaster impacts, advancing scientific research, and guiding urban planning. Precise seismic waveform analysis enables accurate event detection, magnitude estimation, and a deeper understanding of earthquake mechanisms. In this paper, we propose EffiSeisM, an efficient multi-task deep learning model that combines a state space model with a convolutional architecture for earthquake detection, phase picking, and magnitude estimation. EffiSeisM designed a novel Seismic Scale Conv Module and a Conv-SSM encoder, which effectively capture key seismic features while reducing computational complexity. This design ensures high accuracy and operational efficiency, enabling effective seismic analysis. We evaluate EffiSeisM on the DiTing Dataset and DiTing Dataset 2.0, comprising 3 million seismic samples from China and surrounding regions, and compare its performance with several baseline models. The results show that EffiSeisM consistently outperforms the baselines, achieving F1 scores of 0.98 for earthquake detection, 0.92 for phase-P picking, 0.84 for phase-S picking, and an R<span><math><msup><mrow></mrow><mrow><mn>2</mn></mrow></msup></math></span> of 0.92 for magnitude estimation. Additionally, EffiSeisM demonstrates significant improvements in inference speed and accuracy, highlighting its potential as a scalable and efficient solution for large-scale seismic data analysis.</div></div>","PeriodicalId":55221,"journal":{"name":"Computers & Geosciences","volume":"206 ","pages":"Article 106039"},"PeriodicalIF":4.4,"publicationDate":"2025-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145004001","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Short-term earthquake forecasting using electromagnetic and geoacoustic observations via contrastive learning 利用对比学习的电磁和地声观测进行短期地震预报

IF 4.4 2区地球科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Geosciences

Pub Date : 2025-08-19 DOI: 10.1016/j.cageo.2025.106024

Yufeng Jiang, Zining Yu, Haiyong Zheng

Different observations provide earthquake-related information from various perspectives, and effectively leveraging them is essential for enhancing forecasting. Existing data-driven methods primarily rely on concatenation, directly aligning features (e.g., the absolute mean) extracted from different observations side by side. However, this naive approach ignores that each observation reflects a different aspect of the same physical process and inadequately explores cross-observation interactions. To address these issues, we propose CL4EF, a contrastive learning framework that leverages electromagnetic and geoacoustic observations for earthquake forecasting. Specifically, we introduce the synchronous response consistency hypothesis, assuming that different observations within the same time window should respond consistently to the same physical process. Following this hypothesis, we design a contrastive loss that attracts observation pairs from the same station and time window (positives) and repulses others (negatives), enabling cross-observation interaction modeling for the downstream forecasting task. Experimental results demonstrate that CL4EF achieves state-of-the-art performance, improving AUC by 22%. The spatial distribution of forecast probabilities reveals alignment with active fault zones, suggesting the model’s capacity to extract meaningful information for earthquake forecasting. As a result, this study contributes a scalable approach for integrating heterogeneous observations in geosciences and offers new insights into short-term earthquake forecasting.

不同的观测从不同的角度提供了与地震有关的信息，有效地利用这些信息对于加强预报至关重要。现有的数据驱动方法主要依赖于串联，直接对齐从不同观测中提取的特征（例如，绝对平均值）。然而，这种幼稚的方法忽略了每个观测反映了同一物理过程的不同方面，并且没有充分探索交叉观测的相互作用。为了解决这些问题，我们提出了CL4EF，这是一个利用电磁和地声观测进行地震预报的对比学习框架。具体来说，我们引入了同步响应一致性假设，假设同一时间窗口内的不同观测值对同一物理过程的响应一致。根据这一假设，我们设计了一个对比损失，吸引来自同一站和时间窗口的观测对（正），排斥其他观测对（负），从而为下游预测任务实现跨观测交互建模。实验结果表明，CL4EF达到了最先进的性能，使AUC提高了22%。预测概率的空间分布与活动断裂带一致，表明该模型能够提取有意义的地震预报信息。因此，本研究为整合地球科学中的异质观测提供了一种可扩展的方法，并为短期地震预报提供了新的见解。

{"title":"Short-term earthquake forecasting using electromagnetic and geoacoustic observations via contrastive learning","authors":"Yufeng Jiang, Zining Yu, Haiyong Zheng","doi":"10.1016/j.cageo.2025.106024","DOIUrl":"10.1016/j.cageo.2025.106024","url":null,"abstract":"<div><div>Different observations provide earthquake-related information from various perspectives, and effectively leveraging them is essential for enhancing forecasting. Existing data-driven methods primarily rely on concatenation, directly aligning features (e.g., the absolute mean) extracted from different observations side by side. However, this naive approach ignores that each observation reflects a different aspect of the same physical process and inadequately explores cross-observation interactions. To address these issues, we propose CL4EF, a contrastive learning framework that leverages electromagnetic and geoacoustic observations for earthquake forecasting. Specifically, we introduce the synchronous response consistency hypothesis, assuming that different observations within the same time window should respond consistently to the same physical process. Following this hypothesis, we design a contrastive loss that attracts observation pairs from the same station and time window (positives) and repulses others (negatives), enabling cross-observation interaction modeling for the downstream forecasting task. Experimental results demonstrate that CL4EF achieves state-of-the-art performance, improving AUC by 22%. The spatial distribution of forecast probabilities reveals alignment with active fault zones, suggesting the model’s capacity to extract meaningful information for earthquake forecasting. As a result, this study contributes a scalable approach for integrating heterogeneous observations in geosciences and offers new insights into short-term earthquake forecasting.</div></div>","PeriodicalId":55221,"journal":{"name":"Computers & Geosciences","volume":"206 ","pages":"Article 106024"},"PeriodicalIF":4.4,"publicationDate":"2025-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144879268","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Large-scale 3-D magnetotelluric modeling in anisotropic media using extrapolation multigrid method on staggered grids 交错网格外推多网格法在各向异性介质中的大尺度三维大地电磁模拟

IF 4.4 2区地球科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Geosciences

Pub Date : 2025-08-18 DOI: 10.1016/j.cageo.2025.106019

Jinxuan Wang , Kejia Pan , Hongzhu Cai , Zhengguang Liu , Xu Han , Weiwei Ling

To improve the practicality and efficiency of 3D magnetotelluric (MT) data inversion, developing a 3D MT forward modeling algorithm with low computational cost in terms of time and memory is an important prerequisite. An extrapolation cascadic multigrid (EXCMG) method is developed on rectilinear grids to accelerate the solving process of large linear systems arising from the staggered-grid finite difference (SFD) discretization of Maxwell’s equations. Arbitrary anisotropic conductivity is considered, without adding extra unknowns to the SFD scheme. A new prolongation operator based on global extrapolation and mixed-order interpolation is developed to tackle the issue caused by non-nested unknown distribution. The divergence correction scheme for arbitrary anisotropy is employed to stabilize the smoothing process, especially for low-frequency cases. Several examples are tested to validate the accuracy and efficiency of the proposed algorithm, including synthetic models with anisotropy and topography, and the real-world Cascadia model. Results show that our EXCMG solver is more efficient than traditional iterative solvers (e.g., the preconditioned BiCGStab), the algebraic multigrid method and the geometric multigrid (GMG) method. The proposed method can efficiently solve large-scale problems with large grid stretching factors and arbitrary anisotropy, providing powerful engine for large-scale MT inversion.

为了提高大地电磁数据三维反演的实用性和效率，开发一种计算成本低、存储时间短的大地电磁三维正演算法是一个重要的前提。为了加速麦克斯韦方程组交错网格有限差分（SFD）离散引起的大型线性方程组的求解过程，在直线网格上提出了一种外推叶栅多重网格（EXCMG）方法。考虑任意各向异性电导率，而不向SFD方案添加额外的未知数。针对非嵌套未知分布的问题，提出了一种基于全局外推和混合阶内插的扩展算子。采用任意各向异性的散度校正方案来稳定平滑过程，特别是在低频情况下。通过对具有各向异性和地形的合成模型以及实际的Cascadia模型进行了测试，验证了该算法的准确性和效率。结果表明，我们的EXCMG求解器比传统的迭代求解器（如预处理的BiCGStab）、代数多重网格法和几何多重网格法（GMG）更有效。该方法可以有效地解决具有大网格拉伸因子和任意各向异性的大规模问题，为大规模大地电磁法反演提供了强大的引擎。

{"title":"Large-scale 3-D magnetotelluric modeling in anisotropic media using extrapolation multigrid method on staggered grids","authors":"Jinxuan Wang , Kejia Pan , Hongzhu Cai , Zhengguang Liu , Xu Han , Weiwei Ling","doi":"10.1016/j.cageo.2025.106019","DOIUrl":"10.1016/j.cageo.2025.106019","url":null,"abstract":"<div><div>To improve the practicality and efficiency of 3D magnetotelluric (MT) data inversion, developing a 3D MT forward modeling algorithm with low computational cost in terms of time and memory is an important prerequisite. An extrapolation cascadic multigrid (EXCMG) method is developed on rectilinear grids to accelerate the solving process of large linear systems arising from the staggered-grid finite difference (SFD) discretization of Maxwell’s equations. Arbitrary anisotropic conductivity is considered, without adding extra unknowns to the SFD scheme. A new prolongation operator based on global extrapolation and mixed-order interpolation is developed to tackle the issue caused by non-nested unknown distribution. The divergence correction scheme for arbitrary anisotropy is employed to stabilize the smoothing process, especially for low-frequency cases. Several examples are tested to validate the accuracy and efficiency of the proposed algorithm, including synthetic models with anisotropy and topography, and the real-world Cascadia model. Results show that our EXCMG solver is more efficient than traditional iterative solvers (e.g., the preconditioned BiCGStab), the algebraic multigrid method and the geometric multigrid (GMG) method. The proposed method can efficiently solve large-scale problems with large grid stretching factors and arbitrary anisotropy, providing powerful engine for large-scale MT inversion.</div></div>","PeriodicalId":55221,"journal":{"name":"Computers & Geosciences","volume":"206 ","pages":"Article 106019"},"PeriodicalIF":4.4,"publicationDate":"2025-08-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144886346","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Exploiting global information and local edge detail for full waveform inversion 利用全局信息和局部边缘细节进行全波形反演

IF 4.4 2区地球科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Geosciences

Pub Date : 2025-08-14 DOI: 10.1016/j.cageo.2025.106028

Yu-Mei Wang , Qiong Xu , Ziyu Qin , Shulin Pan , Fan Min

Data-driven deep learning full waveform direct inversion (DL-FWI) has emerged as an advanced technique for predicting subsurface structures. Popular approaches frequently encounter blurry edge pixels and inaccurate velocity values. Here, we propose an algorithm called TU-Net that captures both global information and local edge detail to address these issues. With respect to the network design, we incorporate a texture warping module (TWM) into the skip connections of the U-Net backbone. Due to the multi-scale feature extraction ability of TWM, our network is able to learn details in complex regions. With respect to the loss function design, we introduce the mixed pixel and edge (MPE) loss, which is a combination of the mean absolute error, the mean square error, and the edge-based losses. The newly proposed loss function balances the model’s focus on global pixel features with the local edge characterization, driving the network to produce high-quality edges. We apply the proposed approach on publicly available OpenFWI, SEG salt and Marmousi II datasets. Quantitative results demonstrate that TU-Net achieves better performance in terms of MSE, MAE, LPIPS, PSNR, UIQ, and SSIM than four state-of-the-art deep networks. The source code is available at github.com/fansmale/TU-Net.

数据驱动的深度学习全波形直接反演（DL-FWI）已经成为预测地下结构的先进技术。常用的方法经常遇到模糊的边缘像素和不准确的速度值。在这里，我们提出了一种称为TU-Net的算法，它可以捕获全局信息和局部边缘细节来解决这些问题。在网络设计方面，我们将纹理翘曲模块（TWM）集成到U-Net骨干网的跳接中。由于TWM的多尺度特征提取能力，我们的网络能够学习到复杂区域的细节。在损失函数设计方面，我们引入了混合像素和边缘（MPE）损失，它是平均绝对误差、均方误差和基于边缘的损失的组合。新提出的损失函数平衡了模型对全局像素特征和局部边缘特征的关注，推动网络产生高质量的边缘。我们将提出的方法应用于公开可用的OpenFWI、SEG salt和Marmousi II数据集。定量结果表明，TU-Net在MSE、MAE、LPIPS、PSNR、UIQ和SSIM方面的性能优于四种最先进的深度网络。源代码可从github.com/fansmale/TU-Net获得。

{"title":"Exploiting global information and local edge detail for full waveform inversion","authors":"Yu-Mei Wang , Qiong Xu , Ziyu Qin , Shulin Pan , Fan Min","doi":"10.1016/j.cageo.2025.106028","DOIUrl":"10.1016/j.cageo.2025.106028","url":null,"abstract":"<div><div>Data-driven deep learning full waveform direct inversion (DL-FWI) has emerged as an advanced technique for predicting subsurface structures. Popular approaches frequently encounter blurry edge pixels and inaccurate velocity values. Here, we propose an algorithm called TU-Net that captures both global information and local edge detail to address these issues. With respect to the network design, we incorporate a texture warping module (TWM) into the skip connections of the U-Net backbone. Due to the multi-scale feature extraction ability of TWM, our network is able to learn details in complex regions. With respect to the loss function design, we introduce the mixed pixel and edge (MPE) loss, which is a combination of the mean absolute error, the mean square error, and the edge-based losses. The newly proposed loss function balances the model’s focus on global pixel features with the local edge characterization, driving the network to produce high-quality edges. We apply the proposed approach on publicly available OpenFWI, SEG salt and Marmousi II datasets. Quantitative results demonstrate that TU-Net achieves better performance in terms of MSE, MAE, LPIPS, PSNR, UIQ, and SSIM than four state-of-the-art deep networks. The source code is available at github.com/fansmale/TU-Net.</div></div>","PeriodicalId":55221,"journal":{"name":"Computers & Geosciences","volume":"206 ","pages":"Article 106028"},"PeriodicalIF":4.4,"publicationDate":"2025-08-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144852424","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Physics-Informed Fourier-DeepONet for a generalized eikonal solution 广义eikonal解的物理-知情傅里叶-深度网络

IF 4.4 2区地球科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Geosciences

Pub Date : 2025-08-14 DOI: 10.1016/j.cageo.2025.106026

Zhuofan Liu , Goodluck Archibong , Umair Bin Waheed , Sifan Wang , Chao Song

The accurate calculation of seismic traveltime based on the eikonal equation has numerous applications in geophysics, such as microseismic localization and tomography. With the advancement of deep learning, the emergence of neural operators has enabled neural networks to learn general solutions to partial differential equations (PDEs). Moreover, Physics-Informed Neural Network (PINN) allows deep learning models to learn under the supervision of PDEs rather than relying solely on training labels. In this context, we propose utilizing a hybrid model that combines the Deep Operator Network (DeepONet) with the Fourier Neural Operator (FNO) to simulate seismic traveltime under the guidance of eikonal equation, thereby yielding a general solution. We refer to this approach as the Physics-Informed Fourier-DeepONet (PI-Fourier-DeepONet). The loss function of the eikonal equation is calculated by finite difference scheme. We evaluate this method across four different types of seismic structures, and the results demonstrate that PI-Fourier-DeepONet is applicable to a wide range of complex geological structures.

基于eikonal方程的地震走时精确计算在微地震定位和层析成像等地球物理学中有着广泛的应用。随着深度学习的发展，神经算子的出现使得神经网络能够学习偏微分方程（PDEs）的一般解。此外，物理信息神经网络（PINN）允许深度学习模型在pde的监督下学习，而不是仅仅依赖于训练标签。在这种情况下，我们提出利用深度算子网络（DeepONet）和傅立叶神经算子（FNO）相结合的混合模型，在eikonal方程的指导下模拟地震走时，从而得出一般解。我们将这种方法称为物理信息傅里叶-深度网络（PI-Fourier-DeepONet）。对角方程的损失函数采用有限差分格式计算。我们在4种不同类型的地震构造中对该方法进行了评估，结果表明PI-Fourier-DeepONet适用于广泛的复杂地质构造。

{"title":"Physics-Informed Fourier-DeepONet for a generalized eikonal solution","authors":"Zhuofan Liu , Goodluck Archibong , Umair Bin Waheed , Sifan Wang , Chao Song","doi":"10.1016/j.cageo.2025.106026","DOIUrl":"10.1016/j.cageo.2025.106026","url":null,"abstract":"<div><div>The accurate calculation of seismic traveltime based on the eikonal equation has numerous applications in geophysics, such as microseismic localization and tomography. With the advancement of deep learning, the emergence of neural operators has enabled neural networks to learn general solutions to partial differential equations (PDEs). Moreover, Physics-Informed Neural Network (PINN) allows deep learning models to learn under the supervision of PDEs rather than relying solely on training labels. In this context, we propose utilizing a hybrid model that combines the Deep Operator Network (DeepONet) with the Fourier Neural Operator (FNO) to simulate seismic traveltime under the guidance of eikonal equation, thereby yielding a general solution. We refer to this approach as the Physics-Informed Fourier-DeepONet (PI-Fourier-DeepONet). The loss function of the eikonal equation is calculated by finite difference scheme. We evaluate this method across four different types of seismic structures, and the results demonstrate that PI-Fourier-DeepONet is applicable to a wide range of complex geological structures.</div></div>","PeriodicalId":55221,"journal":{"name":"Computers & Geosciences","volume":"206 ","pages":"Article 106026"},"PeriodicalIF":4.4,"publicationDate":"2025-08-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144865465","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Research on hyperspectral remote sensing alteration mineral mapping using an improved ViT model 基于改进ViT模型的高光谱遥感蚀变矿物填图研究

IF 4.4 2区地球科学 Q1 COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS

Computers & Geosciences

Pub Date : 2025-08-09 DOI: 10.1016/j.cageo.2025.106037

Xinhang Feng , Jiejun Huang , Ximing Chen , Han Zhou , Ming Zhang , Chuan Zhang , Fawang Ye

The distribution of altered minerals is a key indicator for finding strategic minerals such as uranium, cobalt, nickel, copper and zinc. In recent years, deep learning has shown outstanding advantages in the field of hyperspectral altered mineral mapping. However, constructing a large volume of high-quality training samples remains time-consuming and labor-intensive. Moreover, many models suffer from limited generalization capability—performing well on training data but exhibiting significant performance degradation on test datasets or in real-world applications. Therefore, a semi-automatic sample construction method was proposed. The sample construction involves three steps. Firstly, using mixed pixel decomposition to extract mineral abundance, then screening samples via mixed matching, and finally enhancing classification accuracy with spectral characteristic quantification. Experimental results show that the test accuracy of the dataset generated by the semi-automated method on the ViT model reached 92.81 %, which is close to that of manually labeled samples at 93.29 %. In terms of models, an improved Vision Transformer (ViT) model was proposed. The SpecPool-Transformer model (SPT) integrates the Grouped Spectral Embedding Module (GSE) and the Convolution-Pooling Module (CPM) to enhance the extraction of adjacent band features from the spectral curves. Additionally, the model's application to cross-source data was achieved through transfer learning. On the SASI dataset of the Baiyanghe uranium deposit, the overall accuracy (OA) and average accuracy (AA) of SpecPool-Transformer reached 96.76 % and 95.14 %, respectively, representing improvements of 3.95 % and 6.11 % over the original ViT model. In the generalization test, the proposed method achieved an OA of 86.10 % and an AA of 83.74 % on the SASI aerial dataset No.1007, outperforming the second-best model, LightGBM, by 20.22 % and 31.15 %, respectively. Field validation results further confirm the high reliability of the proposed model in large-scale alteration mineral mapping across data sources, making it suitable for rapid and extensive alteration mineral mapping applications.

蚀变矿物的分布是寻找诸如铀、钴、镍、铜和锌等战略矿物的关键指标。近年来，深度学习在高光谱蚀变矿物填图领域显示出突出的优势。然而，构建大量高质量的训练样本仍然是耗时和劳动密集型的。此外，许多模型的泛化能力有限——在训练数据上表现良好，但在测试数据集或实际应用中表现出明显的性能下降。为此，提出了一种半自动取样方法。示例构建包括三个步骤。首先利用混合像元分解提取矿物丰度，然后通过混合匹配筛选样品，最后利用光谱特征量化提高分类精度。实验结果表明，半自动化方法在ViT模型上生成的数据集的测试准确率达到92.81%，接近人工标记样本的93.29%。在模型方面，提出了一种改进的视觉变换器（ViT）模型。SpecPool-Transformer模型（SPT）集成了分组频谱嵌入模块（GSE）和卷积池化模块（CPM），增强了对光谱曲线邻带特征的提取。此外，通过迁移学习实现了模型对跨源数据的应用。在白洋河铀矿床SASI数据集上，SpecPool-Transformer模型的总体精度（OA）和平均精度（AA）分别达到96.76%和95.14%，比原ViT模型分别提高了3.95%和6.11%。在SASI航空数据集No.1007的泛化测试中，该方法的OA为86.10%，AA为83.74%，分别比第二优模型LightGBM高出20.22%和31.15%。现场验证结果进一步证实了该模型在跨数据源的大规模蚀变矿物填图中的高可靠性，适用于快速、广泛的蚀变矿物填图应用。

{"title":"Research on hyperspectral remote sensing alteration mineral mapping using an improved ViT model","authors":"Xinhang Feng , Jiejun Huang , Ximing Chen , Han Zhou , Ming Zhang , Chuan Zhang , Fawang Ye","doi":"10.1016/j.cageo.2025.106037","DOIUrl":"10.1016/j.cageo.2025.106037","url":null,"abstract":"<div><div>The distribution of altered minerals is a key indicator for finding strategic minerals such as uranium, cobalt, nickel, copper and zinc. In recent years, deep learning has shown outstanding advantages in the field of hyperspectral altered mineral mapping. However, constructing a large volume of high-quality training samples remains time-consuming and labor-intensive. Moreover, many models suffer from limited generalization capability—performing well on training data but exhibiting significant performance degradation on test datasets or in real-world applications. Therefore, a semi-automatic sample construction method was proposed. The sample construction involves three steps. Firstly, using mixed pixel decomposition to extract mineral abundance, then screening samples via mixed matching, and finally enhancing classification accuracy with spectral characteristic quantification. Experimental results show that the test accuracy of the dataset generated by the semi-automated method on the ViT model reached 92.81 %, which is close to that of manually labeled samples at 93.29 %. In terms of models, an improved Vision Transformer (ViT) model was proposed. The SpecPool-Transformer model (SPT) integrates the Grouped Spectral Embedding Module (GSE) and the Convolution-Pooling Module (CPM) to enhance the extraction of adjacent band features from the spectral curves. Additionally, the model's application to cross-source data was achieved through transfer learning. On the SASI dataset of the Baiyanghe uranium deposit, the overall accuracy (OA) and average accuracy (AA) of SpecPool-Transformer reached 96.76 % and 95.14 %, respectively, representing improvements of 3.95 % and 6.11 % over the original ViT model. In the generalization test, the proposed method achieved an OA of 86.10 % and an AA of 83.74 % on the SASI aerial dataset No.1007, outperforming the second-best model, LightGBM, by 20.22 % and 31.15 %, respectively. Field validation results further confirm the high reliability of the proposed model in large-scale alteration mineral mapping across data sources, making it suitable for rapid and extensive alteration mineral mapping applications.</div></div>","PeriodicalId":55221,"journal":{"name":"Computers & Geosciences","volume":"206 ","pages":"Article 106037"},"PeriodicalIF":4.4,"publicationDate":"2025-08-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144865464","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"地球科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0