Article
Environmental Sciences
Amin Rahimi Dalkhani, Xin Zhang, Cornelis Weemstra
Summary: Seismic travel time tomography using surface waves is an effective tool for three-dimensional crustal imaging, historically based on active seismic sources or earthquakes and more recently on seismic interferometry. Different inversion algorithms are employed, with a newer non-linear tomographic algorithm showing promise in recovering velocity structures accurately in high-density station areas but with some limitations in lower density areas.
Article
Physics, Multidisciplinary
Paulo Douglas S. de Lima, Gilberto Corso, Mauro S. Ferreira, Joao M. de Araujo
Summary: Full-Waveform Inversion (FWI) is a high-resolution technique used in geophysics to evaluate physical parameters and construct subsurface models. The ill-posed nature of FWI presents a challenging problem as multiple models can match the observations. Solving FWI requires efficient sampling techniques to infer parameter information and estimate uncertainties in high-dimensional model spaces. This study investigates the feasibility of using the Hamiltonian Monte Carlo (HMC) method in acoustic FWI and proposes a new strategy for tuning the mass matrix based on seismic survey geometry. The methodology significantly improves the ability of the HMC method in reconstructing reasonable seismic models with reduced computational efforts.
PHYSICA A-STATISTICAL MECHANICS AND ITS APPLICATIONS
(2023)
Article
Geochemistry & Geophysics
Saule Simute, Christian Boehm, Lion Krischer, Alexey Gokhberg, Martin Vallee, Andreas Fichtner
Summary: This study presents probabilistic centroid-moment tensor solutions using a combination of Hamiltonian Monte Carlo sampling and 3-D full-waveform inversion, with a focus on the Japanese islands. The results demonstrate the importance of considering 3-D Earth structure in estimating earthquake parameters, particularly for shorter-period data.
JOURNAL OF GEOPHYSICAL RESEARCH-SOLID EARTH
(2023)
Article
Geochemistry & Geophysics
Xin Zhang, Andrew Curtis
Summary: This study introduces invertible neural networks (INNs) as an alternative to solving nonlinear and nonunique inverse problems in geophysics. By including data uncertainties as additional model parameters and training the network by maximizing the likelihood of the training data, INNs can provide comparable posterior probability density functions to Monte Carlo methods, including correlations between parameters.
JOURNAL OF GEOPHYSICAL RESEARCH-SOLID EARTH
(2021)
Article
Geosciences, Multidisciplinary
Andrea Zunino, Alessandro Ghirotto, Egidio Armadillo, Andreas Fichtner
Summary: This study presents a probabilistic strategy for solving the inverse problem of two-dimensional gravity and magnetic anomaly modeling. It accurately infers possible configurations of geological structures in the subsurface and provides uncertainty estimation.
GEOPHYSICAL RESEARCH LETTERS
(2022)
Article
Chemistry, Physical
Philipp Hoellmer, A. C. Maggs, Werner Krauth
Summary: Event-chain Monte Carlo algorithms for hard-disk dipoles in two dimensions are benchmarked for potential applications in modeling water molecules. The rotation dynamics of dipoles are characterized through integrated autocorrelation times of polarization. The non-reversible event-driven ECMC algorithms show significant speedups compared to the Metropolis algorithm, with differences in speed observed among ECMC variants, indicating Newtonian ECMC as a promising solution for overcoming dynamical arrest in dipolar models with Coulomb interactions.
JOURNAL OF CHEMICAL PHYSICS
(2022)
Article
Computer Science, Information Systems
Wilson Tsakane Mongwe, Rendani Mbuvha, Tshilidzi Marwala
Summary: Hybrid Monte Carlo (HMC) is commonly used in machine learning and statistics; it faces two main practical issues which are addressed through different methods; the A-S2HMC algorithm combines the advantages of NUTS and S2HMC, showing improved performance.
Article
Geochemistry & Geophysics
Ronghua Peng, Bo Han, Yajun Liu, Xiangyun Hu
Summary: A quantitative assessment of model parameter uncertainty is crucial for a reliable interpretation of electromagnetic data. The Bayesian inference framework provides an effective way to rigorously estimate parameter uncertainty related to the recovered solution. The adaptive nature of the reversible jump Markov chain Monte Carlo algorithm allows for inferring the appropriate level of model complexity and associated parameter uncertainty through data.
Article
Geochemistry & Geophysics
Rashed Poormirzaee, Babak Sohrabian, Pejman Tahmasebi
Summary: This article introduces a new inversion framework for seismic refraction traveltime inversion using a hybrid committee artificial neural network and the flower pollination optimization algorithm. Synthetic models are used for training the machine-learning model, and the results are applied in the flower pollination algorithm to estimate the final model. The proposed approach shows successful performance on synthetic and real data.
Article
Geosciences, Multidisciplinary
Wenping Jiang, Ross C. Brodie, Jingming Duan, Ian Roach, Neil Symington, Anandaroop Ray, James Goodwin
Summary: We have developed a Bayesian inference algorithm and made the source code publicly available for the 1D inversion of audio-frequency magnetotelluric data. The algorithm uses a trans-dimensional Markov chain Monte Carlo technique to determine a probabilistic resistivity-depth model. This approach provides a comprehensive exploration of model space and offers a more robust estimation of uncertainty compared to deterministic methods.
JOURNAL OF APPLIED GEOPHYSICS
(2023)
Article
Mathematics, Interdisciplinary Applications
Alexander Buchholz, Nicolas Chopin, Pierre E. Jacob
Summary: Sequential Monte Carlo (SMC) samplers serve as an alternative to MCMC in Bayesian computation, with their performance strongly dependent on the Markov kernels used. The study explores how to automatically calibrate Hamiltonian Monte Carlo kernels within SMC using current particles, building upon the adaptive SMC approach of Fearnhead and Taylor (2013) while also suggesting alternative methods. The advantages of using HMC kernels within an SMC sampler are illustrated through an extensive numerical study.
Article
Computer Science, Artificial Intelligence
Chikezie Chimere Onyekwena, Qi Li, Happiness Ijeoma Umeobi, Xiaying Li, John N. Ng'ombe
Summary: The soil water retention curve is a fundamental concept in unsaturated soil mechanics. Existing prediction models require large datasets, but obtaining such datasets can be costly and time-consuming. To address this issue, a reliability-based approach using a Bayesian framework and Hamiltonian Monte Carlo method is proposed. The method is robust and provides reliable predictions even with limited prior knowledge, while significantly reducing computation time and cost.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Geochemistry & Geophysics
Laura Petrescu, Mihaela Popa, Mircea Radulian
Summary: This study uses the HVSR method to determine the resonance frequency in the Vrancea Seismic Zone in southeast Romania. The results show a correlation between resonance frequency and local surface geology. These findings have important implications for improving seismic hazard estimates and reducing seismic risk.
Article
Computer Science, Interdisciplinary Applications
B. S. Kronheim, M. P. Kuchera, H. B. Prosper
Summary: TensorBNN is a new package that implements Bayesian inference for modern neural network models based on TensorFlow, sampling the posterior density of model parameters using Hamiltonian Monte Carlo. It leverages TensorFlow's architecture and GPU utilization in both training and prediction stages.
COMPUTER PHYSICS COMMUNICATIONS
(2022)
Article
Engineering, Civil
Julien Bonnel, Stan E. Dosso, John A. Goff, Ying-Tsong Lin, James H. Miller, Gopu R. Potty, Preston S. Wilson, David P. Knobles
Summary: This article introduces a transdimensional geoacoustic inversion method for range-dependent propagation tracks, successfully applied to experimental data.
IEEE JOURNAL OF OCEANIC ENGINEERING
(2022)
Article
Geochemistry & Geophysics
Arnab Dhara, Mrinal K. Sen, Reetam Biswas
Summary: We propose a methodology for seismic inversion that generates high-resolution models of facies and elastic properties from pre-stack data. Our algorithm treats the number of layers as unknown and uses a transdimensional approach. We extend the reversible jump Markov Chain Monte Carlo algorithm to simultaneously invert for discrete facies and continuous elastic reservoir properties, considering the non-Gaussian and multimodal nature of model parameters. The integration of facies classification within the inversion improves convergence speed and produces geologically consistent results.
GEOPHYSICAL PROSPECTING
(2023)