Pub Date : 2025-06-01DOI: 10.1016/j.acags.2025.100249
Gaëtan Cortes, Joaquin Garcia-Suarez
In this concise contribution, it is demonstrated that Recurrent Neural Networks (RNNs) based on Gated Recurrent Unit (GRU) architecture, possess the capability to learn the complex dynamics of rate-and-state friction (RSF) laws from synthetic data. The data employed for training the network is generated through the application of traditional RSF equations coupled with either the aging law or the slip law for state evolution. A novel aspect of this approach is the formulation of a loss function that explicitly accounts for the direct effect by means of automatic differentiation. It is found that the GRU-based RNNs effectively learns to predict changes in the friction coefficient resulting from velocity jumps (with and without noise in the target data), thereby showcasing the potential of machine learning models in capturing and simulating the physics of frictional processes. Current limitations and challenges are discussed.
{"title":"Data-driven dynamic friction models based on Recurrent Neural Networks","authors":"Gaëtan Cortes, Joaquin Garcia-Suarez","doi":"10.1016/j.acags.2025.100249","DOIUrl":"10.1016/j.acags.2025.100249","url":null,"abstract":"<div><div>In this concise contribution, it is demonstrated that Recurrent Neural Networks (RNNs) based on Gated Recurrent Unit (GRU) architecture, possess the capability to learn the complex dynamics of rate-and-state friction (RSF) laws from synthetic data. The data employed for training the network is generated through the application of traditional RSF equations coupled with either the aging law or the slip law for state evolution. A novel aspect of this approach is the formulation of a loss function that explicitly accounts for the direct effect by means of automatic differentiation. It is found that the GRU-based RNNs effectively learns to predict changes in the friction coefficient resulting from velocity jumps (with and without noise in the target data), thereby showcasing the potential of machine learning models in capturing and simulating the physics of frictional processes. Current limitations and challenges are discussed.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"26 ","pages":"Article 100249"},"PeriodicalIF":2.6,"publicationDate":"2025-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144243508","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 : 2025-05-21DOI: 10.1016/j.acags.2025.100254
Sungil Kim , Tea-Woo Kim , Yongjun Hong , Hoonyoung Jeong
Accurate carbon dioxide (CO2) phase prediction at the bottomhole of injection wells is essential for ensuring safe and efficient CO2 storage and enhanced gas recovery (EGR). Phase misclassification can cause operational inefficiencies, equipment failure, and compromised storage integrity, posing significant risks to CO2 injection projects. While previous studies have contributed to CO2 phase prediction, they have overlooked well geometry effects, which can impact reliability in real-world applications. This study addresses these challenges by introducing a deep learning framework based on the adaptive factorization network (AFN), which enhances CO2 phase prediction accuracy by leveraging feature interactions. The AFN model was trained on ∼43,000 wells across seven major North American shale gas basins, covering a wide range of well geometries and injection conditions. CO2 phases were classified into supercritical and dense categories, reflecting prevailing flow conditions. To enhance practical applicability, we incorporated real-field wellbore data, ensuring alignment with actual injection environments. The standard AFN model achieved an F1-score of 0.94, with data augmentation further improving performance by reducing false predictions by 50 % and increasing the F1-score to 0.97. Rigorous validation demonstrated the model's robustness for optimizing wellhead temperature to achieve the desired CO2 phase transition. By explicitly considering well geometry effects and real-field conditions, this study advances data-driven CO2 injection modeling, providing a scalable, high-accuracy framework for evaluating CO2 storage and EGR feasibility.
{"title":"Prediction of carbon dioxide phase at bottomhole by adaptive factorization network considering well geometry","authors":"Sungil Kim , Tea-Woo Kim , Yongjun Hong , Hoonyoung Jeong","doi":"10.1016/j.acags.2025.100254","DOIUrl":"10.1016/j.acags.2025.100254","url":null,"abstract":"<div><div>Accurate carbon dioxide (CO<sub>2</sub>) phase prediction at the bottomhole of injection wells is essential for ensuring safe and efficient CO<sub>2</sub> storage and enhanced gas recovery (EGR). Phase misclassification can cause operational inefficiencies, equipment failure, and compromised storage integrity, posing significant risks to CO<sub>2</sub> injection projects. While previous studies have contributed to CO<sub>2</sub> phase prediction, they have overlooked well geometry effects, which can impact reliability in real-world applications. This study addresses these challenges by introducing a deep learning framework based on the adaptive factorization network (AFN), which enhances CO<sub>2</sub> phase prediction accuracy by leveraging feature interactions. The AFN model was trained on ∼43,000 wells across seven major North American shale gas basins, covering a wide range of well geometries and injection conditions. CO<sub>2</sub> phases were classified into supercritical and dense categories, reflecting prevailing flow conditions. To enhance practical applicability, we incorporated real-field wellbore data, ensuring alignment with actual injection environments. The standard AFN model achieved an F1-score of 0.94, with data augmentation further improving performance by reducing false predictions by 50 % and increasing the F1-score to 0.97. Rigorous validation demonstrated the model's robustness for optimizing wellhead temperature to achieve the desired CO<sub>2</sub> phase transition. By explicitly considering well geometry effects and real-field conditions, this study advances data-driven CO<sub>2</sub> injection modeling, providing a scalable, high-accuracy framework for evaluating CO<sub>2</sub> storage and EGR feasibility.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"26 ","pages":"Article 100254"},"PeriodicalIF":2.6,"publicationDate":"2025-05-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144123572","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 : 2025-05-16DOI: 10.1016/j.acags.2025.100252
Birhan Getachew Tikuye, Ram Lakhan Ray
Soil is critical in global carbon storage, holding more carbon than terrestrial vegetation and the atmosphere combined. Accurate soil organic carbon (SOC) estimation is essential for improving agricultural productivity and mitigating climate change. This study aims to explore the retrieval of SOC using a machine learning (ML) approach, leveraging remote sensing data and environmental covariates, focusing on the Lower Brazos River Watershed, southern Texas, USA. The study used Sentinel 2A satellite data-derived indices such as vegetation and water indices, topographic features, soil properties, and climatic factors. Three ML models, namely Gradient Boosting (GB), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost), were deployed, with performance assessed using the R2, RMSE, and MAE. All explanatory variables are geospatial gridded datasets, except for the point-based measurement of SOC on the Prairie View A&M University (PVAMU) research farm plot used to train the model. The RF model demonstrated the best performance in model testing, with the lowest root mean square error (RMSE = 4.17) and mean absolute error (MAE = 3), as well as the highest coefficient of determination (R2 = 0.78). GB was the second-best performing model, achieving an RMSE of 4.23 and an MAE of 3.12, with similar R2 values to the RF model. The average SOC throughout the watershed is 45.5 tons/ha, while the total amount of SOC in the watershed is around 4,278,263 tons. These results suggest that integrating satellite data with environmental covariates and machine learning models holds excellent potential for SOC prediction and supports climate change mitigation efforts by improving carbon stock assessments.
{"title":"Soil organic carbon retrieval using a machine learning approach from satellite and environmental covariates in the Lower Brazos River Watershed, Texas, USA","authors":"Birhan Getachew Tikuye, Ram Lakhan Ray","doi":"10.1016/j.acags.2025.100252","DOIUrl":"10.1016/j.acags.2025.100252","url":null,"abstract":"<div><div>Soil is critical in global carbon storage, holding more carbon than terrestrial vegetation and the atmosphere combined. Accurate soil organic carbon (SOC) estimation is essential for improving agricultural productivity and mitigating climate change. This study aims to explore the retrieval of SOC using a machine learning (ML) approach, leveraging remote sensing data and environmental covariates, focusing on the Lower Brazos River Watershed, southern Texas, USA. The study used Sentinel 2A satellite data-derived indices such as vegetation and water indices, topographic features, soil properties, and climatic factors. Three ML models, namely Gradient Boosting (GB), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost), were deployed, with performance assessed using the R<sup>2</sup>, RMSE, and MAE. All explanatory variables are geospatial gridded datasets, except for the point-based measurement of SOC on the Prairie View A&M University (PVAMU) research farm plot used to train the model. The RF model demonstrated the best performance in model testing, with the lowest root mean square error (RMSE = 4.17) and mean absolute error (MAE = 3), as well as the highest coefficient of determination (R<sup>2</sup> = 0.78). GB was the second-best performing model, achieving an RMSE of 4.23 and an MAE of 3.12, with similar R<sup>2</sup> values to the RF model. The average SOC throughout the watershed is 45.5 tons/ha, while the total amount of SOC in the watershed is around 4,278,263 tons. These results suggest that integrating satellite data with environmental covariates and machine learning models holds excellent potential for SOC prediction and supports climate change mitigation efforts by improving carbon stock assessments.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"26 ","pages":"Article 100252"},"PeriodicalIF":2.6,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144117054","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 : 2025-05-16DOI: 10.1016/j.acags.2025.100251
Jack W. Fekete , Glenn R. Sharman , Xiao Huang
The prodigious use of detrital zircon U-Pb geochronology for provenance studies in recent decades has led many researchers to amass extensive datasets (>100,000 dates). When displayed as age distributions, individual samples are traditionally compared using visual inspection and statistical methods, which can become time-consuming and challenging when using large datasets. We propose that machine learning (ML) can more efficiently classify a sample by its source using detrital zircon U-Pb age distributions. Specifically, we hypothesize that automated machine learning (AutoML), which optimizes algorithm selection and hyperparameters, will outperform an unoptimized Random Forest (RF) classifier and the cross-correlation coefficient (R2), a commonly used metric for comparing age distributions. We test this approach using a well-constrained synthetic dataset and a natural dataset from the Jurassic-Eocene North American Cordillera. In synthetic experiments, AutoML models effectively classify samples by their sources when inter-source similarity across few sources is low to moderate and samples have more than ∼50 analyses. However, the effectiveness of AutoML is highly dependent on sample size and the variability of age modes within the data. Applied to the North American Cordillera dataset, AutoML achieves an ∼0.91 F1 score when predicting between foreland and forearc basin tectonic settings and an ∼0.71 F1 score when predicting subbasins within these settings, outperforming both RF and R2. Moreover, AutoML identifies discriminating age populations between groups, with the average feature importance of 100 models highlighting the 145–125 Ma age range, corresponding to a magmatic lull of the Cordilleran magmatic arc. These results demonstrate AutoML's potential as a powerful predictive and interpretive tool in detrital zircon studies.
{"title":"Classifying detrital zircon U-Pb age distributions using automated machine learning","authors":"Jack W. Fekete , Glenn R. Sharman , Xiao Huang","doi":"10.1016/j.acags.2025.100251","DOIUrl":"10.1016/j.acags.2025.100251","url":null,"abstract":"<div><div>The prodigious use of detrital zircon U-Pb geochronology for provenance studies in recent decades has led many researchers to amass extensive datasets (>100,000 dates). When displayed as age distributions, individual samples are traditionally compared using visual inspection and statistical methods, which can become time-consuming and challenging when using large datasets. We propose that machine learning (ML) can more efficiently classify a sample by its source using detrital zircon U-Pb age distributions. Specifically, we hypothesize that automated machine learning (AutoML), which optimizes algorithm selection and hyperparameters, will outperform an unoptimized Random Forest (RF) classifier and the cross-correlation coefficient (R<sup>2</sup>), a commonly used metric for comparing age distributions. We test this approach using a well-constrained synthetic dataset and a natural dataset from the Jurassic-Eocene North American Cordillera. In synthetic experiments, AutoML models effectively classify samples by their sources when inter-source similarity across few sources is low to moderate and samples have more than ∼50 analyses. However, the effectiveness of AutoML is highly dependent on sample size and the variability of age modes within the data. Applied to the North American Cordillera dataset, AutoML achieves an ∼0.91 F<sub>1</sub> score when predicting between foreland and forearc basin tectonic settings and an ∼0.71 F<sub>1</sub> score when predicting subbasins within these settings, outperforming both RF and R<sup>2</sup>. Moreover, AutoML identifies discriminating age populations between groups, with the average feature importance of 100 models highlighting the 145–125 Ma age range, corresponding to a magmatic lull of the Cordilleran magmatic arc. These results demonstrate AutoML's potential as a powerful predictive and interpretive tool in detrital zircon studies.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"26 ","pages":"Article 100251"},"PeriodicalIF":2.6,"publicationDate":"2025-05-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144131307","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 : 2025-05-15DOI: 10.1016/j.acags.2025.100245
J. Araújo , F. López , S. Johansson , A. Westman , M. Bodin
Incoherent scatter radar (ISR) techniques provide reliable measurements for the analysis of ionospheric plasma. Recent developments in ISR technologies allow the generation of high-resolution 3D data. Examples of such technologies employ the so-called phased-array antenna systems like the AMISR systems in North America or the upcoming EISCAT_3D in the Northern Fennoscandia region. EISCAT_3D will be capable of generating the highest resolution ISR datasets that have ever been measured. We present a novel fast computational strategy for the generation of high-resolution and smooth volumetric ionospheric images that represent ISR data. Through real-time processing, our computational framework will enable a fast decision-making during the monitoring process, where the experimental parameters are adapted in real time as the radars monitor specific phenomena. Real-time monitoring would allow the radar beams to be conveniently pointed at regions of interest and would therefore increase the science impact. We describe our strategy, which implements a flexible mesh generator along with an efficient interpolator specialized for ISR technologies. The proposed strategy is generic in the sense that it can be applied to a large variety of data sets and supports interactive visual analysis and exploration of ionospheric data, supplemented by interactive data transformations and filters.
{"title":"Efficient computation and visualization of ionospheric volumetric images for the enhanced interpretation of Incoherent scatter radar data","authors":"J. Araújo , F. López , S. Johansson , A. Westman , M. Bodin","doi":"10.1016/j.acags.2025.100245","DOIUrl":"10.1016/j.acags.2025.100245","url":null,"abstract":"<div><div>Incoherent scatter radar (ISR) techniques provide reliable measurements for the analysis of ionospheric plasma. Recent developments in ISR technologies allow the generation of high-resolution 3D data. Examples of such technologies employ the so-called phased-array antenna systems like the AMISR systems in North America or the upcoming EISCAT_3D in the Northern Fennoscandia region. EISCAT_3D will be capable of generating the highest resolution ISR datasets that have ever been measured. We present a novel fast computational strategy for the generation of high-resolution and smooth volumetric ionospheric images that represent ISR data. Through real-time processing, our computational framework will enable a fast decision-making during the monitoring process, where the experimental parameters are adapted in real time as the radars monitor specific phenomena. Real-time monitoring would allow the radar beams to be conveniently pointed at regions of interest and would therefore increase the science impact. We describe our strategy, which implements a flexible mesh generator along with an efficient interpolator specialized for ISR technologies. The proposed strategy is generic in the sense that it can be applied to a large variety of data sets and supports interactive visual analysis and exploration of ionospheric data, supplemented by interactive data transformations and filters.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"26 ","pages":"Article 100245"},"PeriodicalIF":2.6,"publicationDate":"2025-05-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144099748","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 : 2025-05-12DOI: 10.1016/j.acags.2025.100242
Sona Salehian Ghamsari , Tonie van Dam , Jack S. Hale
Due to geological features such as fractures, some aquifers demonstrate strongly anisotropic hydraulic behavior. The goal of this study is to use a poroelastic model to calculate surface displacements given known pumping rates to predict the potential utility of Interferometric Synthetic Aperture Radar (InSAR) data for inferring information about anisotropic hydraulic conductivity (AHC) in aquifer systems. To this end, we develop a three-dimensional anisotropic poroelastic model mimicking the main features of the 1994 Anderson Junction aquifer test in southwestern Utah with a 24 to 1 ratio of hydraulic conductivity along the principal axes, previously estimated in the literature using traditional well observation techniques. Under suitable model assumptions, our results show that anisotropy in the hydraulic problem leads to a distinctive elliptical surface displacement pattern centered around the pumping well that could be detected with InSAR. We interpret these results in the context of InSAR acquisition constraints and provide guidelines for designing future pumping tests so that InSAR data can be used to its full potential for improving the characterization of aquifers with anisotropic hydraulic behavior.
{"title":"Can the anisotropic hydraulic conductivity of an aquifer be determined using surface displacement data? A case study","authors":"Sona Salehian Ghamsari , Tonie van Dam , Jack S. Hale","doi":"10.1016/j.acags.2025.100242","DOIUrl":"10.1016/j.acags.2025.100242","url":null,"abstract":"<div><div>Due to geological features such as fractures, some aquifers demonstrate strongly anisotropic hydraulic behavior. The goal of this study is to use a poroelastic model to calculate surface displacements given known pumping rates to predict the potential utility of Interferometric Synthetic Aperture Radar (InSAR) data for inferring information about anisotropic hydraulic conductivity (AHC) in aquifer systems. To this end, we develop a three-dimensional anisotropic poroelastic model mimicking the main features of the 1994 Anderson Junction aquifer test in southwestern Utah with a 24 to 1 ratio of hydraulic conductivity along the principal axes, previously estimated in the literature using traditional well observation techniques. Under suitable model assumptions, our results show that anisotropy in the hydraulic problem leads to a distinctive elliptical surface displacement pattern centered around the pumping well that could be detected with InSAR. We interpret these results in the context of InSAR acquisition constraints and provide guidelines for designing future pumping tests so that InSAR data can be used to its full potential for improving the characterization of aquifers with anisotropic hydraulic behavior.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"26 ","pages":"Article 100242"},"PeriodicalIF":2.6,"publicationDate":"2025-05-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144069086","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}
This study applies machine learning to detect and classify anomalous minerals within a large mineralogical dataset, enhancing geological exploration and resource identification. Using Isolation Forest and One-Class SVM, we identified rare minerals with distinct physical and chemical properties that deviate from common mineral compositions. These anomalies were further grouped using KMeans clustering into three categories, each linked to different geological formation environments: evaporitic, metamorphic, and magmatic processes. The study also evaluates the reliability of these machine learning models using a statistical benchmark and explores the role of deep learning in improving anomaly detection. The findings demonstrate the potential of unsupervised learning to enhance mineral classification, reduce exploration costs, and improve predictive modeling for rare mineral deposits. Future research will refine these methods by integrating Deep Isolation Forest, Autoencoders, and Graph Neural Networks, further strengthening machine learning applications in geosciences.
{"title":"Prediction of rare and anomalous minerals using anomaly detection and machine learning techniques","authors":"Abish Sharapatov , Alisher Saduov , Nazerke Assirbek , Madiyar Abdyrov , Beibit Zhumabayev","doi":"10.1016/j.acags.2025.100250","DOIUrl":"10.1016/j.acags.2025.100250","url":null,"abstract":"<div><div>This study applies machine learning to detect and classify anomalous minerals within a large mineralogical dataset, enhancing geological exploration and resource identification. Using Isolation Forest and One-Class SVM, we identified rare minerals with distinct physical and chemical properties that deviate from common mineral compositions. These anomalies were further grouped using KMeans clustering into three categories, each linked to different geological formation environments: evaporitic, metamorphic, and magmatic processes. The study also evaluates the reliability of these machine learning models using a statistical benchmark and explores the role of deep learning in improving anomaly detection. The findings demonstrate the potential of unsupervised learning to enhance mineral classification, reduce exploration costs, and improve predictive modeling for rare mineral deposits. Future research will refine these methods by integrating Deep Isolation Forest, Autoencoders, and Graph Neural Networks, further strengthening machine learning applications in geosciences.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"26 ","pages":"Article 100250"},"PeriodicalIF":2.6,"publicationDate":"2025-05-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143937834","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 : 2025-05-08DOI: 10.1016/j.acags.2025.100243
Frederik Alexander Falk, Anders Vest Christiansen, Thomas Mejer Hansen
Accurate, efficient, and accessible forward modeling of geophysical processes is essential for understanding them and for inversion of geophysical data. Various algorithms are available for predicting data with the time domain electromagnetic method (TDEM). These algorithms differ in their approach and implementation, making some more suitable than others for specific applications. In this study, we compare three different algorithms for calculating the solution to the 1D forward response problem in TDEM, provided by Geoscience Australia, AarhusInv and SimPEG. Our comparison focuses on four main aspects: efficiency, accuracy, generality and convenience. Efficiency is evaluated from the perspective of computational speed. Accuracy is evaluated in two steps. First, we analyze the relative modeling error of each algorithm’s forward calculation for conductive half-space models, compared to an analytic solution. Secondly, we evaluate the accuracy of the algorithms relative to each other in the context of more complex earth models where no analytic solutions exist. This evaluation assumes a realistic TDEM instrument. Generality is the ability to model a variety of real TDEM scenarios. Lastly, we assess the convenience of each algorithm by considering factors such as ease of use, extensibility, code accessibility, and licensing requirements. We find that no single tested forward algorithm is best for all cases. AarhusInv is accurate and fast while it also has the most options for modeling real TDEM systems, but it requires a license, and is the hardest forward algorithm to interface to. SimPEG is open source, fast, easy to install and results may easily be shared, but has accuracy limitations at early times when modeling real systems with gate integration and low-pass filters. Lastly, Geoscience Australia is open source, accurate, and fast, but can only model dipole sources.
{"title":"Comparison of three one-dimensional time-domain electromagnetic forward algorithms","authors":"Frederik Alexander Falk, Anders Vest Christiansen, Thomas Mejer Hansen","doi":"10.1016/j.acags.2025.100243","DOIUrl":"10.1016/j.acags.2025.100243","url":null,"abstract":"<div><div>Accurate, efficient, and accessible forward modeling of geophysical processes is essential for understanding them and for inversion of geophysical data. Various algorithms are available for predicting data with the time domain electromagnetic method (TDEM). These algorithms differ in their approach and implementation, making some more suitable than others for specific applications. In this study, we compare three different algorithms for calculating the solution to the 1D forward response problem in TDEM, provided by Geoscience Australia, AarhusInv and SimPEG. Our comparison focuses on four main aspects: efficiency, accuracy, generality and convenience. Efficiency is evaluated from the perspective of computational speed. Accuracy is evaluated in two steps. First, we analyze the relative modeling error of each algorithm’s forward calculation for conductive half-space models, compared to an analytic solution. Secondly, we evaluate the accuracy of the algorithms relative to each other in the context of more complex earth models where no analytic solutions exist. This evaluation assumes a realistic TDEM instrument. Generality is the ability to model a variety of real TDEM scenarios. Lastly, we assess the convenience of each algorithm by considering factors such as ease of use, extensibility, code accessibility, and licensing requirements. We find that no single tested forward algorithm is best for all cases. AarhusInv is accurate and fast while it also has the most options for modeling real TDEM systems, but it requires a license, and is the hardest forward algorithm to interface to. SimPEG is open source, fast, easy to install and results may easily be shared, but has accuracy limitations at early times when modeling real systems with gate integration and low-pass filters. Lastly, Geoscience Australia is open source, accurate, and fast, but can only model dipole sources.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"26 ","pages":"Article 100243"},"PeriodicalIF":2.6,"publicationDate":"2025-05-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143937833","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 : 2025-05-07DOI: 10.1016/j.acags.2025.100247
Amad Hussen, Tanveer Alam Munshi, Minhaz Chowdhury, Labiba Nusrat Jahan, Abu Bakker Siddique, Mahamudul Hashan
Characterizing reservoir rock is aided by an understanding of how the permeability changes dynamically within formations. There are currently only nine research articles that focus on permeability prediction using a small set of input parameters that are easily, affordably, and frequently derived from laboratory core analysis. The majority of machine learning models applied to permeability determination are connected to well logs. This work investigates and implements four novel approaches for permeability prediction from standard core analysis data. These approaches include hybrid stacking and three boosting techniques: AdaBoost, gradient boosting, and extreme gradient boosting (XGB). While boosting enhances any regressor or classifier by being computationally efficient in large-scale datasets, stacking increases prediction accuracy by mixing the output from several base models. The dataset comprises measures of porosity (), grain density (), water saturation (), oil saturation (), depth, and absolute permeability () for 197 core plugs from the sedimentary basin of Jeanne d'Arc. According to the results, boosting strategies with a root mean squared error (RMSE) of less than 32.24 and a coefficient of determination (R2) of more than 0.95 are good enough and meet the study's objectives. With an RMSE of 23.45–30.16 and an R2 of 0.92–0.95, hybrid stacking—which combines AdaBoost, gradient boosting, XGB, and artificial neural networks (ANN)— offers a bit less accuracy than boosting models. Gradient Boosting is shown to provide the maximum precision, with an RMSE of 18.23 and an R2 of 0.98. The ANN has also high prediction accuracy, with an R2 of 0.97 and an RMSE of 26.41. The boosting strategies in permeability prediction from routine core data are quite accurate, as shown by the comparison of the suggested methodology with 20 earlier utilized models identified in 9 literatures. XGB, Gradient Boosting, AdaBoost, and Stacking models are explored in this study, marking the first instance of their application in predicting permeability from routine core analysis. Additionally, previously utilized algorithms, such as ANN, have also been re-evaluated to predict permeability. All proposed algorithms are systematically ranked based on performance criteria. The models developed in this research, leveraging a few key inputs, offer engineers and scientists a reliable and efficient means of determining reservoir permeability with high accuracy. This significantly reduces the reliance on resource-intensive and time-consuming laboratory analyses.
{"title":"Evaluating reservoir permeability from core data: Leveraging boosting techniques and ANN for heterogeneous reservoirs","authors":"Amad Hussen, Tanveer Alam Munshi, Minhaz Chowdhury, Labiba Nusrat Jahan, Abu Bakker Siddique, Mahamudul Hashan","doi":"10.1016/j.acags.2025.100247","DOIUrl":"10.1016/j.acags.2025.100247","url":null,"abstract":"<div><div>Characterizing reservoir rock is aided by an understanding of how the permeability changes dynamically within formations. There are currently only nine research articles that focus on permeability prediction using a small set of input parameters that are easily, affordably, and frequently derived from laboratory core analysis. The majority of machine learning models applied to permeability determination are connected to well logs. This work investigates and implements four novel approaches for permeability prediction from standard core analysis data. These approaches include hybrid stacking and three boosting techniques: AdaBoost, gradient boosting, and extreme gradient boosting (XGB). While boosting enhances any regressor or classifier by being computationally efficient in large-scale datasets, stacking increases prediction accuracy by mixing the output from several base models. The dataset comprises measures of porosity (<span><math><mrow><mo>∅</mo></mrow></math></span>), grain density (<span><math><mrow><msub><mi>ρ</mi><mrow><mi>g</mi><mi>r</mi></mrow></msub></mrow></math></span>), water saturation (<span><math><mrow><msub><mi>S</mi><mi>W</mi></msub></mrow></math></span>), oil saturation (<span><math><mrow><msub><mi>S</mi><mi>O</mi></msub></mrow></math></span>), depth, and absolute permeability (<span><math><mrow><mi>K</mi></mrow></math></span>) for 197 core plugs from the sedimentary basin of Jeanne d'Arc. According to the results, boosting strategies with a root mean squared error (RMSE) of less than 32.24 and a coefficient of determination (R<sup>2</sup>) of more than 0.95 are good enough and meet the study's objectives. With an RMSE of 23.45–30.16 and an R<sup>2</sup> of 0.92–0.95, hybrid stacking—which combines AdaBoost, gradient boosting, XGB, and artificial neural networks (ANN)— offers a bit less accuracy than boosting models. Gradient Boosting is shown to provide the maximum precision, with an RMSE of 18.23 and an R<sup>2</sup> of 0.98. The ANN has also high prediction accuracy, with an R2 of 0.97 and an RMSE of 26.41. The boosting strategies in permeability prediction from routine core data are quite accurate, as shown by the comparison of the suggested methodology with 20 earlier utilized models identified in 9 literatures. XGB, Gradient Boosting, AdaBoost, and Stacking models are explored in this study, marking the first instance of their application in predicting permeability from routine core analysis. Additionally, previously utilized algorithms, such as ANN, have also been re-evaluated to predict permeability. All proposed algorithms are systematically ranked based on performance criteria. The models developed in this research, leveraging a few key inputs, offer engineers and scientists a reliable and efficient means of determining reservoir permeability with high accuracy. This significantly reduces the reliance on resource-intensive and time-consuming laboratory analyses.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"26 ","pages":"Article 100247"},"PeriodicalIF":2.6,"publicationDate":"2025-05-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143937832","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 : 2025-05-04DOI: 10.1016/j.acags.2025.100246
M.E. Al-Atroush
Accurately estimating the axial capacity of drilled shafts remains a persistent challenge in geotechnical engineering, as evidenced by significant discrepancies between measured load-test results and theoretical predictions. To bridge this gap, a novel Deep Learning–Physics-Informed Neural Network (DL-PINN) framework is proposed. Employing Meyerhof's bearing capacity equations as a physics-based constraint, the developed PINN integrated soil and geometric parameters directly into its training loss function. By combining these first-principles relationships with empirical data, the model preserved fundamental geotechnical mechanisms while refining predictive accuracy through dynamic weight adjustments between data-driven and physics-based loss components. Comparative experiments with a standard artificial neural network (ANN), using a dataset derived from the loaded-to-failure in-situ pile test and subsequent numerical simulations, demonstrated that although the ANN may attain lower statistical errors, the PINN's adherence to physical laws yields predictions that better align with established geotechnical behavior. This balance between physics fidelity and data adaptability may nominate these PINN frameworks to address the “black box” nature of deep learning in geotechnical applications. The paper also suggested the future research needs to fulfill the scientific and practical gap.
{"title":"A deep learning physics-informed neural network (PINN) for predicting drilled shaft axial capacity","authors":"M.E. Al-Atroush","doi":"10.1016/j.acags.2025.100246","DOIUrl":"10.1016/j.acags.2025.100246","url":null,"abstract":"<div><div>Accurately estimating the axial capacity of drilled shafts remains a persistent challenge in geotechnical engineering, as evidenced by significant discrepancies between measured load-test results and theoretical predictions. To bridge this gap, a novel Deep Learning–Physics-Informed Neural Network (DL-PINN) framework is proposed. Employing Meyerhof's bearing capacity equations as a physics-based constraint, the developed PINN integrated soil and geometric parameters directly into its training loss function. By combining these first-principles relationships with empirical data, the model preserved fundamental geotechnical mechanisms while refining predictive accuracy through dynamic weight adjustments between data-driven and physics-based loss components. Comparative experiments with a standard artificial neural network (ANN), using a dataset derived from the loaded-to-failure in-situ pile test and subsequent numerical simulations, demonstrated that although the ANN may attain lower statistical errors, the PINN's adherence to physical laws yields predictions that better align with established geotechnical behavior. This balance between physics fidelity and data adaptability may nominate these PINN frameworks to address the “black box” nature of deep learning in geotechnical applications. The paper also suggested the future research needs to fulfill the scientific and practical gap.</div></div>","PeriodicalId":33804,"journal":{"name":"Applied Computing and Geosciences","volume":"26 ","pages":"Article 100246"},"PeriodicalIF":2.6,"publicationDate":"2025-05-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143917318","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