{"id":{"repo_id":"texas-state","oai_identifier":"oai:digital.library.txst.edu:10877/20495"},"canonical_url":"https://search.dev.ndltd.org/etd/texas-state/oai:digital.library.txst.edu:10877/20495","repository":{"repo_id":"texas-state","name":"Texas State University","base_url":"https://digital.library.txst.edu/server/oai/request"},"display":{"title":"Geo-Informed Deep Learning for Spatial Downscaling of Solute Transport in Heterogeneous Porous Media","abstract":"Resolving solute transport in heterogeneous porous media is a complex phenomena as it encounters a data sparsity challenge when investigated experimentally and a computational cost challenge when simulated numerically. This work proposes a unique two-stage deep learning architecture comprising a dual-branch autoencoder and a Geo-informed super-resolution generative adversarial network (Gi-SRGAN) to address this dual challenge. The dual-branch autoencoder addresses the issue of sparsity by constructing a continuous, but coarse representation of concentration and pressure profiles from a sparse, discontinuous profile with up to 85% missing data points. The Gi-SRGAN is then employed to generate a finer representation of field variables from the outputs generated by the dual-branch autoencoder (i.e., downscaling). We train and test our framework using five solute transport cases with varying levels of heterogeneity and compare the results with standalone methods, namely the vanilla autoencoder and vanilla SRGAN in addition to ground truth profiles generated by finite element method (FEM). The comparison are performed based on several statistical metrics such as absolute point error (APE), mean squared error (MSE), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and learned perceptual image patch similarity (LPIPS). The first four cases are used for training and evaluation, while the last case is utilized for blind testing to determine the generalization capability of the framework. Our results show that the dual-branch autoencoder outperforms the vanilla autoencoder, and the Gi-SRGAN outperforms the SRGAN during both the training and evaluation phases. Moreover, the proposed framework can successfully construct the fine representation of concentration profiles, compared to FEM, using coarse representation of pressure, concentration, and domain’s permeability fields. When tested on a blind test case, the dual-branch autoencoder and Gi-SRGAN exhibit superior performance compared to their counterparts in terms of evaluation metrics, while also accurately estimating the ground truth collectively.","abstract_html":"Resolving solute transport in heterogeneous porous media is a complex phenomena as it encounters a data sparsity challenge when investigated experimentally and a computational cost challenge when simulated numerically. This work proposes a unique two-stage deep learning architecture comprising a dual-branch autoencoder and a Geo-informed super-resolution generative adversarial network (Gi-SRGAN) to address this dual challenge. The dual-branch autoencoder addresses the issue of sparsity by constructing a continuous, but coarse representation of concentration and pressure profiles from a sparse, discontinuous profile with up to 85% missing data points. The Gi-SRGAN is then employed to generate a finer representation of field variables from the outputs generated by the dual-branch autoencoder (i.e., downscaling). We train and test our framework using five solute transport cases with varying levels of heterogeneity and compare the results with standalone methods, namely the vanilla autoencoder and vanilla SRGAN in addition to ground truth profiles generated by finite element method (FEM). The comparison are performed based on several statistical metrics such as absolute point error (APE), mean squared error (MSE), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and learned perceptual image patch similarity (LPIPS). The first four cases are used for training and evaluation, while the last case is utilized for blind testing to determine the generalization capability of the framework. Our results show that the dual-branch autoencoder outperforms the vanilla autoencoder, and the Gi-SRGAN outperforms the SRGAN during both the training and evaluation phases. Moreover, the proposed framework can successfully construct the fine representation of concentration profiles, compared to FEM, using coarse representation of pressure, concentration, and domain’s permeability fields. When tested on a blind test case, the dual-branch autoencoder and Gi-SRGAN exhibit superior performance compared to their counterparts in terms of evaluation metrics, while also accurately estimating the ground truth collectively.","abstract_has_math":false,"creators":["Pawar, Nikhil"],"institution":"Texas State University","degree_name":"Master of Science","degree_level":"Masters","degree_discipline":"Engineering","degree_department":null,"school":null,"contributors":[],"advisors":["Faroughi, Salah A."],"committee_chairs":[],"committee_members":["Das, Subasish","Valles, Damian"],"year":2023,"date_issued":"2023-12","date_published":"2023-12","updated_at":"2026-07-27T21:22:39Z","subjects":["data imputation","spatial downscaling","heterogeneous porous media"],"languages":["en"],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/10877/20495","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Faroughi, Salah A."]},{"key":"dc:contributor.committeemember","label":"Committee Member","values":["Das, Subasish","Valles, Damian"]},{"key":"dc:creator","label":"Author","values":["Pawar, Nikhil"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2025-03-11T19:06:09Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2025-03-11T19:06:09Z"]},{"key":"dc:date.issued","label":"Date","values":["2023-12"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Engineering"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Masters"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Master of Science"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["Texas State University"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["data imputation","spatial downscaling","heterogeneous porous media"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/10877/20495"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Resolving solute transport in heterogeneous porous media is a complex phenomena as it encounters a data sparsity challenge when investigated experimentally and a computational cost challenge when simulated numerically. This work proposes a unique two-stage deep learning architecture comprising a dual-branch autoencoder and a Geo-informed super-resolution generative adversarial network (Gi-SRGAN) to address this dual challenge. The dual-branch autoencoder addresses the issue of sparsity by constructing a continuous, but coarse representation of concentration and pressure profiles from a sparse, discontinuous profile with up to 85% missing data points. The Gi-SRGAN is then employed to generate a finer representation of field variables from the outputs generated by the dual-branch autoencoder (i.e., downscaling). We train and test our framework using five solute transport cases with varying levels of heterogeneity and compare the results with standalone methods, namely the vanilla autoencoder and vanilla SRGAN in addition to ground truth profiles generated by finite element method (FEM). The comparison are performed based on several statistical metrics such as absolute point error (APE), mean squared error (MSE), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and learned perceptual image patch similarity (LPIPS). The first four cases are used for training and evaluation, while the last case is utilized for blind testing to determine the generalization capability of the framework. Our results show that the dual-branch autoencoder outperforms the vanilla autoencoder, and the Gi-SRGAN outperforms the SRGAN during both the training and evaluation phases. Moreover, the proposed framework can successfully construct the fine representation of concentration profiles, compared to FEM, using coarse representation of pressure, concentration, and domain’s permeability fields. When tested on a blind test case, the dual-branch autoencoder and Gi-SRGAN exhibit superior performance compared to their counterparts in terms of evaluation metrics, while also accurately estimating the ground truth collectively."]},{"key":"dc:format","label":"Dc Format","values":["Text"]},{"key":"dc:format.medium","label":"Dc Format Medium","values":["1 file (.pdf)"]},{"key":"dc:title","label":"Title","values":["Geo-Informed Deep Learning for Spatial Downscaling of Solute Transport in Heterogeneous Porous Media"]}]}],"canonical_facts":{"dc:contributor.advisor":["Faroughi, Salah A."],"dc:contributor.committeemember":["Das, Subasish","Valles, Damian"],"dc:creator":["Pawar, Nikhil"],"dc:date.accessioned":["2025-03-11T19:06:09Z"],"dc:date.available":["2025-03-11T19:06:09Z"],"dc:date.issued":["2023-12"],"dc:description.abstract":["Resolving solute transport in heterogeneous porous media is a complex phenomena as it encounters a data sparsity challenge when investigated experimentally and a computational cost challenge when simulated numerically. This work proposes a unique two-stage deep learning architecture comprising a dual-branch autoencoder and a Geo-informed super-resolution generative adversarial network (Gi-SRGAN) to address this dual challenge. The dual-branch autoencoder addresses the issue of sparsity by constructing a continuous, but coarse representation of concentration and pressure profiles from a sparse, discontinuous profile with up to 85% missing data points. The Gi-SRGAN is then employed to generate a finer representation of field variables from the outputs generated by the dual-branch autoencoder (i.e., downscaling). We train and test our framework using five solute transport cases with varying levels of heterogeneity and compare the results with standalone methods, namely the vanilla autoencoder and vanilla SRGAN in addition to ground truth profiles generated by finite element method (FEM). The comparison are performed based on several statistical metrics such as absolute point error (APE), mean squared error (MSE), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and learned perceptual image patch similarity (LPIPS). The first four cases are used for training and evaluation, while the last case is utilized for blind testing to determine the generalization capability of the framework. Our results show that the dual-branch autoencoder outperforms the vanilla autoencoder, and the Gi-SRGAN outperforms the SRGAN during both the training and evaluation phases. Moreover, the proposed framework can successfully construct the fine representation of concentration profiles, compared to FEM, using coarse representation of pressure, concentration, and domain’s permeability fields. When tested on a blind test case, the dual-branch autoencoder and Gi-SRGAN exhibit superior performance compared to their counterparts in terms of evaluation metrics, while also accurately estimating the ground truth collectively."],"dc:format":["Text"],"dc:format.medium":["1 file (.pdf)"],"dc:identifier.uri":["https://hdl.handle.net/10877/20495"],"dc:language.iso":["en"],"dc:subject":["data imputation","spatial downscaling","heterogeneous porous media"],"dc:title":["Geo-Informed Deep Learning for Spatial Downscaling of Solute Transport in Heterogeneous Porous Media"],"dc:type":["Thesis"],"thesis:degree_discipline":["Engineering"],"thesis:degree_level":["Masters"],"thesis:degree_name":["Master of Science"],"thesis:institution_name":["Texas State University"]},"updated_at":"2026-07-27T21:22:39Z"}