Three‐Dimensional Permeability Inversion Using Convolutional Neural Networks and Positron Emission Tomography

Huang, Zitong; Kurotori, Takeshi; Pini, Ronny; Benson, Sally M.; Zahasky, Christopher

doi:10.1029/2021wr031554

Cited by 12 publications

(7 citation statements)

References 73 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While bacterial radiolabeling with [ 18 F]-FDG for use in PET imaging has been investigated in clinical settings, geoscientific application of PET has been limited to monitoring flow and transport in porous and fractured media. − In this study, the combination of radiolabelingE. coli with positron-emitting radioisotopes and imaging the dynamic 3D E.…”

Section: Discussionmentioning

confidence: 99%

In Situ Measurements of Dynamic Bacteria Transport and Attachment in Heterogeneous Sand-Packed Columns

Le,

Thompson,

Roden

et al. 2023

Environ. Sci. Technol.

Self Cite

View full text Add to dashboard Cite

Prevention, mitigation, and regulation of bacterial contaminants in groundwater require a fundamental understanding of the mechanisms of transport and attachment in complex geological materials. Discrepancies in bacterial transport behaviors observed between field studies and laboratory experiments indicate an incomplete understanding of dynamic bacterial transport and immobilization processes in realistic heterogeneous geologic systems. Here, we develop a new experimental approach for in situ quantification of dynamic bacterial transport and attachment distribution in geologic media that relies on radiolabelingEscherichia coliwith positron-emitting radioisotopes and quantifying transport with three-dimensional (3D) positron emission tomography (PET) imaging. Our results indicate that the highest bacterial attachment occurred at the interfaces between sand layers oriented orthogonal to the direction of flow. The predicted bacterial attachment from a 3D numerical model matched the experimental PET results, highlighting that the experimentally observed bacterial transport behavior can be accurately captured with a distribution of a first-order irreversible attachment model. This is the first demonstration of the direct measurement of attachment coefficient distributions from bacterial transport experiments in geologic media and provides a transformational approach to better understand bacterial transport mechanisms, improve model parametrization, and accurately predict how local geologic conditions can influence bacterial fate and transport in groundwater.

show abstract

Section: Discussionmentioning

confidence: 99%

In Situ Measurements of Dynamic Bacteria Transport and Attachment in Heterogeneous Sand-Packed Columns

Le,

Thompson,

Roden

et al. 2023

Environ. Sci. Technol.

Self Cite

View full text Add to dashboard Cite

show abstract

“…Inverse modeling of hydraulic tomography (HT) estimates spatially distributed fields of hydrogeological parameters, such as hydraulic conductivity, transmissivity and specific storage, using steady or time‐dependent hydraulic head measurements in sequential pumping tests (Butler Jr. et al., 1999; Cardiff et al., 2009; Gottlieb & Dietrich, 1995; Huang et al., 2022; Illman et al., 2008, 2010; Liu et al., 2013; Liu et al., 2014; Tosaka et al., 1993; Yeh & Lee, 2007; Yeh & Liu, 2000; Zha et al., 2018; Zhu & Yeh, 2005). The most widely used approach for solving HT inverse problems is geostatistical approach (GA), including quasilinear GA (Fienen et al., 2008; Kitanidis, 1995) and successive linear estimator (SLE) (Yeh et al., 1995) The major bottleneck of GA is that it requires iterative forward model simulations to evaluate the Jacobian matrix, which are computationally expensive for large‐scale, high‐dimensional models.…”

Section: Introductionmentioning

confidence: 99%

“…Huang et al. (2022) proposed using CNNs and positron emission tomography (PET) data to estimate the permeability distribution in 3D subsurface domains. The CNN architecture was designed based on 3D convolution and can effectively capture spatial patterns and relationships within the PET data.…”

Section: Introductionmentioning

confidence: 99%

“…However, since the underlying patterns of the channel field are not explicitly encoded, such models cannot be used to generate realizations, nor can they be combined with other inverse methods for estimating high-dimensional parameter fields. Huang et al (2022) proposed using CNNs and positron emission tomography (PET) data to estimate the permeability distribution in 3D subsurface domains. The CNN architecture was designed based on 3D convolution and can effectively capture spatial patterns and relationships within the PET data.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Predictive Deep Learning for High‐Dimensional Inverse Modeling of Hydraulic Tomography in Gaussian and Non‐Gaussian Fields

Guo,

Liu,

Luo

2023

Water Resources Research

View full text Add to dashboard Cite

Inverse modeling of hydraulic tomography (HT) is computationally expensive for estimating high‐dimensional hydrogeologic parameter fields. In this work, we develop a novel method called HT‐INV‐NN, which combines dimensionality reduction techniques with a predictive deep learning (DL) model to estimate high‐dimensional Gaussian and non‐Gaussian channel fields. The HT‐INV‐NN model consists of a predictor that directly learns the inverse process from hydraulic head measurements to latent variables of random fields, and a decoder that generates high‐dimensional parameter fields from predicted latent variables. For Gaussian spatially correlated fields, the decoder utilizes principal components derived from spatial covariance, and for non‐Gaussian channel fields, a generative adversarial network (GAN) is trained using generated realizations based on a training image (TI). The predictor is a deep neural network calibrated using the reference data obtained from HT forward simulations, which can be implemented in parallel. HT‐INV‐NN is successfully tested in multiple numerical experiments including steady‐state and transient HT for estimating Gaussian fields in 2D and 3D, as well as binary discontinuous or continuous non‐Gaussian channel fields. The training process is efficient, and the model structure demonstrates robustness for input data with perturbations. The model performance on multiple validation data sets are satisfactory when compared with other numerical and deep learning methods.

show abstract

“…Specially, (a) porosity indicates the density and storage and controls groundwater flowing (Anovitz & Cole, 2015; Nur et al., 1998). (b) Permeability, and connectivity as a function of pore morphology characterize the flow behaviors for single phase or multiphase subsurface flow (Bernabé et al., 2010; Haj Ibrahim et al., 2019; Huang et al., 2022; Lahiri, 2021). (c) Formation factor indicates water saturation or non‐conductive phase saturation in geophysics (Cai et al., 2017, 2019).…”

Section: Introductionmentioning

confidence: 99%

Predicting 3D Physical Properties From a Single 2D Slice Based on Convolutional Neural Networks: 2D‐Slice‐To‐3D‐Properties for Porous Rocks

Chen,

Yang,

et al. 2023

Water Resources Research

View full text Add to dashboard Cite

The 3D physical properties of porous rocks directly determine the subsurface flow and modelling. However, predicting a wide range of 3D physical rocks remains a formidable challenge and requires a large amount of input. Reliable microstructure‐property correlations can accurately predict 3D physical properties, avoiding time‐consuming experimental testing. Here, we propose a new dimensionless CNN‐based method to find inherent correlations for accurately predicting 3D properties of porous rocks using only a single 2D slice or a series of 2D slices. Training and testing were conducted on 9 properties of 16,409 semi‐realistic 3D porous rocks. Eight properties, except for the formation factor, were predicted with an acceptable correlation coefficient of 0.92. Furthermore, the modeling results indicate that the network architecture of the model has a significant impact on its performance. The classical CNN Model 4 achieved an average validating loss of 1.0×10‐3. Interval 2D slice sampling can significantly reduce the computational cost in training model. The developed model was verified and tested by five independent porous samples. Additionally, the proposed method accurately predicted permeability in different directions using only 2D slices trained in one direction, for porous rocks where anisotropy is not significant. This work demonstrates the ability to predict various 3D properties rapidly and accurately using only a single 2D slice from semi‐realistic 3D volumes. It provides insights into the conversion of 2D slices to 3D properties for porous media in a wide range of applications, including property estimation from 2D images without structural construction and simulation in 3D.This article is protected by copyright. All rights reserved.

show abstract

Three‐Dimensional Permeability Inversion Using Convolutional Neural Networks and Positron Emission Tomography

Cited by 12 publications

References 73 publications

In Situ Measurements of Dynamic Bacteria Transport and Attachment in Heterogeneous Sand-Packed Columns

In Situ Measurements of Dynamic Bacteria Transport and Attachment in Heterogeneous Sand-Packed Columns

Predictive Deep Learning for High‐Dimensional Inverse Modeling of Hydraulic Tomography in Gaussian and Non‐Gaussian Fields

Predicting 3D Physical Properties From a Single 2D Slice Based on Convolutional Neural Networks: 2D‐Slice‐To‐3D‐Properties for Porous Rocks

Contact Info

Product

Resources

About