Pub Date : 2026-01-14DOI: 10.1016/j.aiig.2026.100189
Shuai Lv , Yuxiang Peng
With the rapid development of artificial intelligence in seismology, various deep learning-based seismic phase picking models have emerged in recent years. However, existing models face challenges in balancing picking accuracy with computational efficiency for real-time applications. To address this issue, we propose DTPP, a novel seismic phase picking network that integrates depthwise separable convolution and temporal dilated convolution. The model adopts a backbone-feature fusion-decoder architecture, utilizing depthwise separable convolution and dilated convolution to significantly expand the receptive field while reducing computational complexity. We trained the model on the STEAD dataset and evaluated its performance on the global GEEDataset V1.0(84,782 independent samples after excluding overlapping STEAD data to ensure fair cross-dataset evaluation). Experimental results demonstrate that DTPP achieves a P-wave recall of 0.877, F1 score of 0.878, and average P/S F1 score of 0.714, ranking first among all comparison models. Meanwhile, DTPP maintains high computational efficiency with only 0.25M parameters, 0.98 MB model size, and 3ms single-sample inference time per batch, making it suitable for real-time seismic monitoring applications. The proposed method provides an effective solution to the accuracy-efficiency trade-off problem in seismic phase picking tasks.
{"title":"DTPP:An efficient depthwise separable TCN for seismic phase picking","authors":"Shuai Lv , Yuxiang Peng","doi":"10.1016/j.aiig.2026.100189","DOIUrl":"10.1016/j.aiig.2026.100189","url":null,"abstract":"<div><div>With the rapid development of artificial intelligence in seismology, various deep learning-based seismic phase picking models have emerged in recent years. However, existing models face challenges in balancing picking accuracy with computational efficiency for real-time applications. To address this issue, we propose DTPP, a novel seismic phase picking network that integrates depthwise separable convolution and temporal dilated convolution. The model adopts a backbone-feature fusion-decoder architecture, utilizing depthwise separable convolution and dilated convolution to significantly expand the receptive field while reducing computational complexity. We trained the model on the STEAD dataset and evaluated its performance on the global GEEDataset V1.0(84,782 independent samples after excluding overlapping STEAD data to ensure fair cross-dataset evaluation). Experimental results demonstrate that DTPP achieves a P-wave recall of 0.877, F1 score of 0.878, and average P/S F1 score of 0.714, ranking first among all comparison models. Meanwhile, DTPP maintains high computational efficiency with only 0.25M parameters, 0.98 MB model size, and 3ms single-sample inference time per batch, making it suitable for real-time seismic monitoring applications. The proposed method provides an effective solution to the accuracy-efficiency trade-off problem in seismic phase picking tasks.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"7 1","pages":"Article 100189"},"PeriodicalIF":4.2,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146037971","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-14DOI: 10.1016/j.aiig.2026.100188
Zhiqiang Lan , Yaqi Zhang , Yaojun Wang , Keyu Chen , Haoxiang Yang , Yinzhu Chen , Yangyang Yu
Artificial intelligence-based methods for picking microseismic phase arrivals have been widely adopted. However, these methods are frequently challenged by complex and dynamic monitoring scenarios, where various types of environmental noise mask low-energy microseismic signals. Moreover, the paucity of labelled data often impairs the reliability and accuracy of their results. To address these issues, this study proposes a novel supervised learning framework named FC-Net, which integrates automatic labelling via Fuzzy C-means clustering (FCM) with the U-Net architecture. Specifically, the FCM algorithm is employed to derive the probabilistic distributions of microseismic phase arrival times, which are then used as training labels for model training. The proposed FC-Net is equipped with soft attention gates (AGs) and recurrent-residual convolution units (RRCUs), which effectively enhance the network's ability to focus on key seismic features. The arrival time is determined as the moment when the predicted probability exceeds a predefined threshold for the first arrival pick. Evaluated on a field dataset collected from Southwest China, FC-Net is demonstrated to outperform the conventional U-Net method. The experimental results demonstrate that FC-Net achieves adaptive labeling, enhances the detection rate of microseismic events, and improves the precision of phase arrival picking. Furthermore, it exhibits strong generalization performance across microseismic events with varying signal-to-noise ratios (SNRs).
{"title":"An FCM-based microseismic phase arrival picking method and application","authors":"Zhiqiang Lan , Yaqi Zhang , Yaojun Wang , Keyu Chen , Haoxiang Yang , Yinzhu Chen , Yangyang Yu","doi":"10.1016/j.aiig.2026.100188","DOIUrl":"10.1016/j.aiig.2026.100188","url":null,"abstract":"<div><div>Artificial intelligence-based methods for picking microseismic phase arrivals have been widely adopted. However, these methods are frequently challenged by complex and dynamic monitoring scenarios, where various types of environmental noise mask low-energy microseismic signals. Moreover, the paucity of labelled data often impairs the reliability and accuracy of their results. To address these issues, this study proposes a novel supervised learning framework named FC-Net, which integrates automatic labelling via Fuzzy C-means clustering (FCM) with the U-Net architecture. Specifically, the FCM algorithm is employed to derive the probabilistic distributions of microseismic phase arrival times, which are then used as training labels for model training. The proposed FC-Net is equipped with soft attention gates (AGs) and recurrent-residual convolution units (RRCUs), which effectively enhance the network's ability to focus on key seismic features. The arrival time is determined as the moment when the predicted probability exceeds a predefined threshold for the first arrival pick. Evaluated on a field dataset collected from Southwest China, FC-Net is demonstrated to outperform the conventional U-Net method. The experimental results demonstrate that FC-Net achieves adaptive labeling, enhances the detection rate of microseismic events, and improves the precision of phase arrival picking. Furthermore, it exhibits strong generalization performance across microseismic events with varying signal-to-noise ratios (SNRs).</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"7 1","pages":"Article 100188"},"PeriodicalIF":4.2,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977632","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-02DOI: 10.1016/j.aiig.2026.100185
Ahmed Kotb , Marwa S. Moustafa , Safaa Hassan , Hesham Hassan
Light Detection and Ranging (LIDAR) point clouds provide high precision spatial data but impose significant storage and transmission challenges, often exceeding one gigabyte per square kilometer. This paper introduces a novel hierarchical framework for lossless LiDAR data compression, designed to address these issues through a three-stage approach: class-aware segmentation, adaptive algorithm selection, and hierarchical compression. The framework begins by partitioning point clouds into semantic classes (e.g., ground, vegetation, buildings) using an SVM-based classifier with a radial basis function kernel, enabling targeted compression that exploits intra-class redundancies. The adaptive algorithm selection stage employs a density-based matcher to choose optimal compression algorithms for each class, ensuring efficiency across varying point densities and terrain types. Finally, hierarchical compression merges class-specific compressed files and applies a secondary compression using WinRAR for enhanced efficiency. Evaluated on ten openly available benchmark LiDAR datasets, the proposed method consistently outperforms state-of-the-art lossless compression techniques, such as LASzip, achieving file size reductions to 12.76 % of the original for high-density point clouds and 22.51 % for low-density ones. While compression and decompression times are higher than some alternatives, the framework's superior storage savings and perfect fidelity make it ideal for large-scale LiDAR data archiving and exchange.
{"title":"An adaptable hybrid method for lossless airborne lidar data compression","authors":"Ahmed Kotb , Marwa S. Moustafa , Safaa Hassan , Hesham Hassan","doi":"10.1016/j.aiig.2026.100185","DOIUrl":"10.1016/j.aiig.2026.100185","url":null,"abstract":"<div><div>Light Detection and Ranging (LIDAR) point clouds provide high precision spatial data but impose significant storage and transmission challenges, often exceeding one gigabyte per square kilometer. This paper introduces a novel hierarchical framework for lossless LiDAR data compression, designed to address these issues through a three-stage approach: class-aware segmentation, adaptive algorithm selection, and hierarchical compression. The framework begins by partitioning point clouds into semantic classes (e.g., ground, vegetation, buildings) using an SVM-based classifier with a radial basis function kernel, enabling targeted compression that exploits intra-class redundancies. The adaptive algorithm selection stage employs a density-based matcher to choose optimal compression algorithms for each class, ensuring efficiency across varying point densities and terrain types. Finally, hierarchical compression merges class-specific compressed files and applies a secondary compression using WinRAR for enhanced efficiency. Evaluated on ten openly available benchmark LiDAR datasets, the proposed method consistently outperforms state-of-the-art lossless compression techniques, such as LASzip, achieving file size reductions to 12.76 % of the original for high-density point clouds and 22.51 % for low-density ones. While compression and decompression times are higher than some alternatives, the framework's superior storage savings and perfect fidelity make it ideal for large-scale LiDAR data archiving and exchange.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"7 1","pages":"Article 100185"},"PeriodicalIF":4.2,"publicationDate":"2026-01-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884817","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-30DOI: 10.1016/j.aiig.2025.100184
Iver Martinsen , Benjamin Ricaud , David Wade , Odd Kolbjørnsen , Fred Godtliebsen
Microfossil analysis is important in subsurface mapping, for example to match strata between wells. This analysis is currently conducted by specialist geoscientists who manually investigate large numbers of physical samples with the aim of identifying informative microfossil species and genera. The current digitalization of large volumes of microfossil samples that is being conducted by the Norwegian Offshore Directorate, paired with AI development, opens up new opportunities for automating parts of the analysis to help the geologist in the analysis. Unsupervised representation learning is a research area in Artificial Intelligence (AI) that lies at the core of this challenge, as this way of learning can create useful image representations by utilizing large volumes of data without requiring labels. Previous work has presented good results for the classification of a limited number of classes, but there are still challenges related to classification in realistic settings where additional unknown species are present. In this paper, we connect unsupervised representation learning and uncertainty estimation and create a comprehensive tool to automate microfossil analysis. We present our methodology and results in three parts. In the first part, we train several AI models from scratch using state-of-the-art self-supervised learning methods, obtaining excellent results compared against state-of-the-art foundation models for image classification and content-based image retrieval. In the second part, we develop a method based on conformal prediction which enables our classifier to handle a large pool of images containing both in-distribution and out-of-distribution data, while at the same time allowing us to create error estimates to control the uncertainty of the prediction sets. In the third part, we use our method to create distribution charts of fossils for a range of genera in multiple wells.
{"title":"The Fossil Frontier: An answer to the 3-billion fossil question","authors":"Iver Martinsen , Benjamin Ricaud , David Wade , Odd Kolbjørnsen , Fred Godtliebsen","doi":"10.1016/j.aiig.2025.100184","DOIUrl":"10.1016/j.aiig.2025.100184","url":null,"abstract":"<div><div>Microfossil analysis is important in subsurface mapping, for example to match strata between wells. This analysis is currently conducted by specialist geoscientists who manually investigate large numbers of physical samples with the aim of identifying informative microfossil species and genera. The current digitalization of large volumes of microfossil samples that is being conducted by the Norwegian Offshore Directorate, paired with AI development, opens up new opportunities for automating parts of the analysis to help the geologist in the analysis. Unsupervised representation learning is a research area in Artificial Intelligence (AI) that lies at the core of this challenge, as this way of learning can create useful image representations by utilizing large volumes of data without requiring labels. Previous work has presented good results for the classification of a limited number of classes, but there are still challenges related to classification in realistic settings where additional unknown species are present. In this paper, we connect unsupervised representation learning and uncertainty estimation and create a comprehensive tool to automate microfossil analysis. We present our methodology and results in three parts. In the first part, we train several AI models from scratch using state-of-the-art self-supervised learning methods, obtaining excellent results compared against state-of-the-art foundation models for image classification and content-based image retrieval. In the second part, we develop a method based on conformal prediction which enables our classifier to handle a large pool of images containing both in-distribution and out-of-distribution data, while at the same time allowing us to create error estimates to control the uncertainty of the prediction sets. In the third part, we use our method to create distribution charts of fossils for a range of genera in multiple wells.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"7 1","pages":"Article 100184"},"PeriodicalIF":4.2,"publicationDate":"2025-12-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145939042","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-27DOI: 10.1016/j.aiig.2025.100183
Osama Massarweh , Abdul Salam Abd , Jens Schneider , Ahmad S. Abushaikha
Accurate prediction of reservoir permeability based on geostatistical modeling and history matching is often limited by spatial resolution and computational efficiency. To address this limitation, we developed a novel supervised machine learning (ML) approach employing feedforward neural networks (FFNNs) to predict spatial permeability distribution in heterogeneous carbonate reservoirs from production well rates. The ML model was trained on 25 black oil reservoir simulation cases derived from a geologically realistic representation of the Upper Kharaib Member in the United Arab Emirates. Input features for training included cell spatial coordinates , distances between cells and the closest wells, and corresponding time-weighted oil production rates extracted from simulation outputs for each well. The target output was the permeability at each cell. The grid consisted of 22,739 structured cells, and training scenarios considered different closest well counts ( 1, 5, 10, and 20). The prediction performance of the trained model was evaluated across 12 unseen test cases. The model achieved higher accuracy with increased well input (), demonstrating the potential of ML for efficient permeability estimation. This study highlights the effectiveness of integrating physical simulation outputs and spatial production patterns within a neural network structure for robust reservoir characterization.
{"title":"Application of machine learning for permeability prediction in heterogeneous carbonate reservoirs","authors":"Osama Massarweh , Abdul Salam Abd , Jens Schneider , Ahmad S. Abushaikha","doi":"10.1016/j.aiig.2025.100183","DOIUrl":"10.1016/j.aiig.2025.100183","url":null,"abstract":"<div><div>Accurate prediction of reservoir permeability based on geostatistical modeling and history matching is often limited by spatial resolution and computational efficiency. To address this limitation, we developed a novel supervised machine learning (ML) approach employing feedforward neural networks (FFNNs) to predict spatial permeability distribution in heterogeneous carbonate reservoirs from production well rates. The ML model was trained on 25 black oil reservoir simulation cases derived from a geologically realistic representation of the Upper Kharaib Member in the United Arab Emirates. Input features for training included cell spatial coordinates <span><math><mrow><mo>(</mo><msub><mrow><mi>x</mi></mrow><mrow><mi>i</mi></mrow></msub><mo>,</mo><msub><mrow><mi>y</mi></mrow><mrow><mi>i</mi></mrow></msub><mo>,</mo><msub><mrow><mi>z</mi></mrow><mrow><mi>i</mi></mrow></msub><mo>)</mo></mrow></math></span>, distances between cells and the <span><math><mi>n</mi></math></span> closest wells, and corresponding time-weighted oil production rates extracted from simulation outputs for each well. The target output was the permeability at each cell. The grid consisted of 22,739 structured cells, and training scenarios considered different closest well counts (<span><math><mrow><mi>n</mi><mo>=</mo></mrow></math></span> 1, 5, 10, and 20). The prediction performance of the trained model was evaluated across 12 unseen test cases. The model achieved higher accuracy with increased well input (<span><math><mi>n</mi></math></span>), demonstrating the potential of ML for efficient permeability estimation. This study highlights the effectiveness of integrating physical simulation outputs and spatial production patterns within a neural network structure for robust reservoir characterization.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"7 1","pages":"Article 100183"},"PeriodicalIF":4.2,"publicationDate":"2025-12-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145884816","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-09DOI: 10.1016/j.aiig.2025.100171
Jhonatan Rivera-Rivera , Héctor Aguilera , Marta Béjar-Pizarro , Carolina Guardiola-Albert , Pablo Ezquerro , Anna Barra
Ground deformation processes, such as landslides and subsidence, cause significant social, economic, and environmental impacts. This study aims to automatically classify ground deformation processes in Spain using a machine learning approach applied to InSAR-based datasets. The database integrates InSAR measurement points (MPs) from 20 case studies in Spain, obtained from various institutional sources, and 32 geoenvironmental variables related to ground deformation, morphometry, geology, climate, and land use. The proposed classification strategy follows a hierarchical structure with two levels: first, distinguishing between landslides and subsidence; then, identifying the specific type within each main class (mining landslide, environmental landslide, constructive subsidence, mining subsidence, and piezometric subsidence). Several machine learning algorithms (Naïve Bayes, Logistic Regression, Decision Tree, Random Forest, Extra Trees, Gradient Boosting Machine, XGBoost, LightGBM, and CatBoost) and data configurations were tested, combining different spatial resolutions and class balancing techniques. The best performance (Cohen's Kappa = 0.78) was achieved with the hierarchical approach using the 200 m grid dataset, applying XGBoost for the parental and landslide models, and CatBoost for the subsidence model. Using this approach, 70 % de test sites achieved over 88 % correctly classified cells, 20 % had between 50 % and 83 %, and only one test case was entirely misclassified. The analysis of the most relevant variables indicates that annual mean precipitation, mining activity, buildings, landslide susceptibility, and slope are key factors. These results demonstrate the potential of the hierarchical approach to improve classification and lay the groundwork for future application at national and European scales, incorporating new training cases, process types, and continental data sources. In conclusion, this study presents, for the first time, a hierarchical machine learning model capable of accurately classifying ground deformation processes in Spain, with the aim of supporting territorial management and geohazard mitigation.
地面变形过程，如滑坡和下沉，会造成重大的社会、经济和环境影响。本研究旨在使用应用于基于insar的数据集的机器学习方法对西班牙的地面变形过程进行自动分类。该数据库整合了来自西班牙20个案例研究的InSAR测量点（MPs），这些数据来自不同的机构来源，以及与地面变形、地貌测量、地质、气候和土地利用相关的32个地球环境变量。提出的分类策略遵循两个层次的分层结构：第一，区分滑坡和沉降；然后，在每个主要类别中确定具体类型（采矿滑坡、环境滑坡、建设性沉陷、采矿沉陷和压力沉降）。结合不同的空间分辨率和类平衡技术，测试了几种机器学习算法（Naïve Bayes, Logistic Regression, Decision Tree, Random Forest, Extra Trees, Gradient Boosting machine, XGBoost， LightGBM和CatBoost）和数据配置。使用200米网格数据集的分层方法获得了最佳性能（Cohen’s Kappa = 0.78），对亲代和滑坡模型应用XGBoost，对沉降模型应用CatBoost。使用这种方法，70%的测试站点实现了超过88%的正确分类单元，20%的站点在50%到83%之间，并且只有一个测试用例完全被错误分类。对最相关变量的分析表明，年平均降水、采矿活动、建筑物、滑坡易感性和坡度是关键因素。这些结果显示了层次方法改进分类的潜力，并为将来在国家和欧洲范围内的应用奠定了基础，结合了新的培训案例、过程类型和大陆数据源。总之，本研究首次提出了一种分层机器学习模型，能够准确地对西班牙的地面变形过程进行分类，目的是支持领土管理和减轻地质灾害。
{"title":"Hierarchical machine learning for the automatic classification of surface deformation from SAR observations","authors":"Jhonatan Rivera-Rivera , Héctor Aguilera , Marta Béjar-Pizarro , Carolina Guardiola-Albert , Pablo Ezquerro , Anna Barra","doi":"10.1016/j.aiig.2025.100171","DOIUrl":"10.1016/j.aiig.2025.100171","url":null,"abstract":"<div><div>Ground deformation processes, such as landslides and subsidence, cause significant social, economic, and environmental impacts. This study aims to automatically classify ground deformation processes in Spain using a machine learning approach applied to InSAR-based datasets. The database integrates InSAR measurement points (MPs) from 20 case studies in Spain, obtained from various institutional sources, and 32 geoenvironmental variables related to ground deformation, morphometry, geology, climate, and land use. The proposed classification strategy follows a hierarchical structure with two levels: first, distinguishing between landslides and subsidence; then, identifying the specific type within each main class (mining landslide, environmental landslide, constructive subsidence, mining subsidence, and piezometric subsidence). Several machine learning algorithms (Naïve Bayes, Logistic Regression, Decision Tree, Random Forest, Extra Trees, Gradient Boosting Machine, XGBoost, LightGBM, and CatBoost) and data configurations were tested, combining different spatial resolutions and class balancing techniques. The best performance (Cohen's Kappa = 0.78) was achieved with the hierarchical approach using the 200 m grid dataset, applying XGBoost for the parental and landslide models, and CatBoost for the subsidence model. Using this approach, 70 % de test sites achieved over 88 % correctly classified cells, 20 % had between 50 % and 83 %, and only one test case was entirely misclassified. The analysis of the most relevant variables indicates that annual mean precipitation, mining activity, buildings, landslide susceptibility, and slope are key factors. These results demonstrate the potential of the hierarchical approach to improve classification and lay the groundwork for future application at national and European scales, incorporating new training cases, process types, and continental data sources. In conclusion, this study presents, for the first time, a hierarchical machine learning model capable of accurately classifying ground deformation processes in Spain, with the aim of supporting territorial management and geohazard mitigation.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"7 1","pages":"Article 100171"},"PeriodicalIF":4.2,"publicationDate":"2025-12-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145841219","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-03DOI: 10.1016/j.aiig.2025.100173
Reza Taherdangkoo , Thomas Nagel , Vladimir Tyurin , Chaofan Chen , Faramarz Doulati Ardejani , Christoph Butscher
Soil–water retention (SWR) is fundamental for understanding the hydro-mechanical behavior of unsaturated clay soils. The soil–water retention curve is typically obtained through extensive and costly laboratory testing. To offer a more efficient alternative, an extreme gradient boosting (XGBoost) model, optimized using a hybrid particle swarm optimization and genetic algorithm (PSO–GA), was developed. This hybrid model estimates the SWR across a broad suction range, accounting for both drying and wetting paths, along with key soil parameters. The performance of the model was evaluated through various statistical analyses and by comparing the predicted gravimetric water content with experimental data. A backward feature elimination method was employed to assess the impact of various input parameters on model accuracy and to offer a simplified model for scenarios with limited data availability. Additionally, Monte Carlo simulations were conducted to quantify the inherent uncertainties associated with the dataset, XGBoost hyperparameters, and model performance. The hybrid PSO–GA XGBoost model effectively estimates the water retention of clayey soils during both drying and wetting cycles, proving to be an alternative to traditional soil mechanics correlations.
{"title":"Prediction of the soil–water retention curve of compacted clays using PSO–GA XGBoost","authors":"Reza Taherdangkoo , Thomas Nagel , Vladimir Tyurin , Chaofan Chen , Faramarz Doulati Ardejani , Christoph Butscher","doi":"10.1016/j.aiig.2025.100173","DOIUrl":"10.1016/j.aiig.2025.100173","url":null,"abstract":"<div><div>Soil–water retention (SWR) is fundamental for understanding the hydro-mechanical behavior of unsaturated clay soils. The soil–water retention curve is typically obtained through extensive and costly laboratory testing. To offer a more efficient alternative, an extreme gradient boosting (XGBoost) model, optimized using a hybrid particle swarm optimization and genetic algorithm (PSO–GA), was developed. This hybrid model estimates the SWR across a broad suction range, accounting for both drying and wetting paths, along with key soil parameters. The performance of the model was evaluated through various statistical analyses and by comparing the predicted gravimetric water content with experimental data. A backward feature elimination method was employed to assess the impact of various input parameters on model accuracy and to offer a simplified model for scenarios with limited data availability. Additionally, Monte Carlo simulations were conducted to quantify the inherent uncertainties associated with the dataset, XGBoost hyperparameters, and model performance. The hybrid PSO–GA XGBoost model effectively estimates the water retention of clayey soils during both drying and wetting cycles, proving to be an alternative to traditional soil mechanics correlations.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"7 1","pages":"Article 100173"},"PeriodicalIF":4.2,"publicationDate":"2025-12-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145747577","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01DOI: 10.1016/j.aiig.2025.100172
Rodrigo Uribe-Ventura , Willem Viveen , Ferdinand Pineda-Ancco , César Beltrán-Castañon
Landslides constitute one of the most destructive geological hazards worldwide, causing thousands of casualties and billions in economic losses annually. To mitigate these risks, accurate and efficient pixel-wise mapping of landslides for automatic semantic segmentation is of paramount importance. While recent advances in deep learning, particularly with transformer architectures and large pre-trained models like the Segment Anything Model (SAM), have shown promise, their application to landslide mapping is often hindered by high computational costs, prompt dependency, and challenges with data imbalance. To address these limitations, we propose GeoNeXt, a novel semantic segmentation architecture for intelligent landslide recognition. It combines a scalable, pre-trained ConvNeXt V2 encoder with a decoder that utilizes Pyramid Squeeze Attention (PSA) and Atrous Spatial Pyramid Pooling (ASPP) to capture multi-scale features. Through domain adaptation on the large-scale CAS landslide dataset, we refined the encoder's general pre-trained features to learn robust, landslide-specific features. GeoNeXt exhibited zero-shot transferability, achieving 74–78 % F1 and 64–66 % mIoU across three distinct test datasets from diverse regions, which were entirely excluded from the training process. Ablation studies on decoder variants validated the PSA-ASPP synergy, achieving a superior F1 of 90.39 % and mIoU of 83.18 % on the CAS dataset. Comparative analysis confirmed that GeoNeXt outperformed SAM-based methods, achieving F1 scores of 94.25 %, 86.43 %, and 92.27 % (mIoU: 89.51 %, 78.21 %, 86.02 %) on the Bijie, Landslide4Sense, and GVLM datasets, respectively, with 10× fewer parameters than SAM-based methods and lower computational demands. We showed that modernized convolutions, paired with strategic training, were a viable alternative to resource-intensive transformers. This efficiency facilitated their use in operational intelligent landslide recognition and geohazard monitoring systems.
{"title":"GeoNeXt: Efficient landslide mapping using a pre-trained ConvNeXt V2 encoder with a PSA-ASPP decoder","authors":"Rodrigo Uribe-Ventura , Willem Viveen , Ferdinand Pineda-Ancco , César Beltrán-Castañon","doi":"10.1016/j.aiig.2025.100172","DOIUrl":"10.1016/j.aiig.2025.100172","url":null,"abstract":"<div><div>Landslides constitute one of the most destructive geological hazards worldwide, causing thousands of casualties and billions in economic losses annually. To mitigate these risks, accurate and efficient pixel-wise mapping of landslides for automatic semantic segmentation is of paramount importance. While recent advances in deep learning, particularly with transformer architectures and large pre-trained models like the Segment Anything Model (SAM), have shown promise, their application to landslide mapping is often hindered by high computational costs, prompt dependency, and challenges with data imbalance. To address these limitations, we propose GeoNeXt, a novel semantic segmentation architecture for intelligent landslide recognition. It combines a scalable, pre-trained ConvNeXt V2 encoder with a decoder that utilizes Pyramid Squeeze Attention (PSA) and Atrous Spatial Pyramid Pooling (ASPP) to capture multi-scale features. Through domain adaptation on the large-scale CAS landslide dataset, we refined the encoder's general pre-trained features to learn robust, landslide-specific features. GeoNeXt exhibited zero-shot transferability, achieving 74–78 % F1 and 64–66 % mIoU across three distinct test datasets from diverse regions, which were entirely excluded from the training process. Ablation studies on decoder variants validated the PSA-ASPP synergy, achieving a superior F1 of 90.39 % and mIoU of 83.18 % on the CAS dataset. Comparative analysis confirmed that GeoNeXt outperformed SAM-based methods, achieving F1 scores of 94.25 %, 86.43 %, and 92.27 % (mIoU: 89.51 %, 78.21 %, 86.02 %) on the Bijie, Landslide4Sense, and GVLM datasets, respectively, with 10× fewer parameters than SAM-based methods and lower computational demands. We showed that modernized convolutions, paired with strategic training, were a viable alternative to resource-intensive transformers. This efficiency facilitated their use in operational intelligent landslide recognition and geohazard monitoring systems.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"6 2","pages":"Article 100172"},"PeriodicalIF":4.2,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145683651","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01DOI: 10.1016/j.aiig.2025.100163
Gong Cheng , Syed Hussain , Yingdong Yang , Li Sun , Asad Atta , Cheng Huang , Guangqiang Li , Mohammad Naseer , Lingyi Liao
Recent advancements in remote sensing technology have made it easier to detect surface faults. Deep learning, especially convolutional models, offers new potential for automatic fault detection from remote sensing imagery. However, these models often struggle with segmentation accuracy due to their limitations in handling spatial hierarchies and short-range dependencies. They process data in local contexts, which is insufficient for tasks requiring an understanding of global structures, like fault detection. This leads to inaccurate boundary divisions and incomplete fault trace detections. To address these issues, the Convolution Holographic Reduced Representations-Based Unet (CHRRA-Unet) is introduced. This U-shaped network combines convolution and a novel attention-based transformer for remote sensing image segmentation. By extracting both local and global features, the CHRRA-Unet significantly improves the detection of geological faults in remote sensing images. By incorporating a convolutional module (CM) and holographic reduced representation attention (HRRA), local and global feature extraction is improved. To minimize computational complexity, the traditional Multi-Layer Perceptron (MLP) is replaced with the Local Perception Module (LPM). The Multi-Feature Conversion Module (MFCM) ensures an effective combination of feature maps during encoding and decoding, enhancing the network's ability to accurately detect fault traces. Extensive experiments show that CHRRA-Unet achieves a high accuracy rate of 97.20 % in remote sensing image segmentation, outperforming existing models and providing superior fault detection capabilities over current methods.
{"title":"Enhancing fault detection using CHRRA-Unet and focal loss functions for imbalanced data: A case study in Luoping county, Yunnan, China","authors":"Gong Cheng , Syed Hussain , Yingdong Yang , Li Sun , Asad Atta , Cheng Huang , Guangqiang Li , Mohammad Naseer , Lingyi Liao","doi":"10.1016/j.aiig.2025.100163","DOIUrl":"10.1016/j.aiig.2025.100163","url":null,"abstract":"<div><div>Recent advancements in remote sensing technology have made it easier to detect surface faults. Deep learning, especially convolutional models, offers new potential for automatic fault detection from remote sensing imagery. However, these models often struggle with segmentation accuracy due to their limitations in handling spatial hierarchies and short-range dependencies. They process data in local contexts, which is insufficient for tasks requiring an understanding of global structures, like fault detection. This leads to inaccurate boundary divisions and incomplete fault trace detections. To address these issues, the Convolution Holographic Reduced Representations-Based Unet (CHRRA-Unet) is introduced. This U-shaped network combines convolution and a novel attention-based transformer for remote sensing image segmentation. By extracting both local and global features, the CHRRA-Unet significantly improves the detection of geological faults in remote sensing images. By incorporating a convolutional module (CM) and holographic reduced representation attention (HRRA), local and global feature extraction is improved. To minimize computational complexity, the traditional Multi-Layer Perceptron (MLP) is replaced with the Local Perception Module (LPM). The Multi-Feature Conversion Module (MFCM) ensures an effective combination of feature maps during encoding and decoding, enhancing the network's ability to accurately detect fault traces. Extensive experiments show that CHRRA-Unet achieves a high accuracy rate of 97.20 % in remote sensing image segmentation, outperforming existing models and providing superior fault detection capabilities over current methods.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"7 1","pages":"Article 100163"},"PeriodicalIF":4.2,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145798249","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01DOI: 10.1016/j.aiig.2025.100166
Mohamad Rifai , Harintaka
Tropical lakes such as Lake Sentarum in Kalimantan, Indonesia, represent ecologically rich ecosystems with high biodiversity and constitute the largest lake on the island of Kalimantan. This lake serves as a sensitive indicator of climate change; however, its monitoring is often hindered by persistent cloud cover. This study evaluates the effectiveness of a Gradient Tree Boosting machine learning model integrated with multisource satellite data, including optical imagery, Sentinel-1 SAR, Sentinel-2, and high resolution NICFI data, in accurately mapping surface water dynamics. The Gradient Tree Boosting model was trained and validated using water and non water samples collected from annual imagery spanning 2019 to 2024, achieving validation accuracies ranging from 80 percent to 97 percent. Results demonstrate that Gradient Tree Boosting successfully integrates the strengths of each sensor, producing consistent annual water maps despite extreme hydrological fluctuations caused by El Niño and La Niña events. These findings highlight the model's potential application in water resource management, particularly in providing accurate baseline data to support adaptation planning for droughts and floods in climate vulnerable regions.
印度尼西亚加里曼丹的森塔鲁姆湖（Lake Sentarum）等热带湖泊代表了生态丰富、生物多样性高的生态系统，是加里曼丹岛上最大的湖泊。这个湖是气候变化的敏感指标；然而，它的监测经常受到持续云层覆盖的阻碍。本研究评估了结合多源卫星数据（包括光学图像、Sentinel-1 SAR、Sentinel-2和高分辨率NICFI数据）的Gradient Tree Boosting机器学习模型在精确绘制地表水动力学地图方面的有效性。梯度树增强模型使用从2019年至2024年的年度图像中收集的水和非水样本进行训练和验证，验证精度从80%到97%不等。结果表明，Gradient Tree Boosting成功地整合了每个传感器的优势，尽管El Niño和La Niña事件造成了极端的水文波动，但仍能生成一致的年度水图。这些发现突出了该模型在水资源管理中的潜在应用，特别是在提供准确的基线数据以支持气候脆弱地区干旱和洪水的适应规划方面。
{"title":"Unveiling climate-driven water surface dynamics in the largest tropical lake in Borneo: A machine learning approach using multi-source satellite imagery","authors":"Mohamad Rifai , Harintaka","doi":"10.1016/j.aiig.2025.100166","DOIUrl":"10.1016/j.aiig.2025.100166","url":null,"abstract":"<div><div>Tropical lakes such as Lake Sentarum in Kalimantan, Indonesia, represent ecologically rich ecosystems with high biodiversity and constitute the largest lake on the island of Kalimantan. This lake serves as a sensitive indicator of climate change; however, its monitoring is often hindered by persistent cloud cover. This study evaluates the effectiveness of a Gradient Tree Boosting machine learning model integrated with multisource satellite data, including optical imagery, Sentinel-1 SAR, Sentinel-2, and high resolution NICFI data, in accurately mapping surface water dynamics. The Gradient Tree Boosting model was trained and validated using water and non water samples collected from annual imagery spanning 2019 to 2024, achieving validation accuracies ranging from 80 percent to 97 percent. Results demonstrate that Gradient Tree Boosting successfully integrates the strengths of each sensor, producing consistent annual water maps despite extreme hydrological fluctuations caused by El Niño and La Niña events. These findings highlight the model's potential application in water resource management, particularly in providing accurate baseline data to support adaptation planning for droughts and floods in climate vulnerable regions.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"6 2","pages":"Article 100166"},"PeriodicalIF":4.2,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145623402","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号