MPMichael J. Pyrcz

Michael J. Pyrcz

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Michael J. Pyrcz has a total of 110 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Predicting CO2 Storage and Enhanced Gas Recovery (CSEGR) Performance under Uncertainty using Ensemble-Based Reservoir Heterogeneity Modeling

    by Deena A. Elhossary, Kishore K. Mohanty, Ademide O. Mabadeje, Michael J. Pyrcz

    CO2 can be injected into gas reservoirs for CO2 storage and Enhanced Gas Recovery (CSEGR). The goal of this work is to quantify the impact of reservoir heterogeneity on the performance of CSEGR. An ensemble-based modeling approach is used for quantifying uncertainty in reservoir characterization and performance predictions. We developed a systematic design of experiments to generate and analyze a large ensemble of heterogeneity models coupled with CSEGR-tailored flow simulators. Our proposed workflow includes (1) development of a geostatistical framework for generating multiple reservoir realizations under controlled heterogeneity scenarios, (2) implementation of Monte Carlo sampling with 100 realizations per scenario to quantify subsurface uncertainty, and (3) application of ensemble flow simulations to assess the impact of reservoir heterogeneity on CSEGR key performance metrics including CH4 recovery factor, CO2 storage efficiency, breakthrough time and sweep efficiency. The stochastic heterogeneity models are variogram-based with sequential indicator simulation, to generate multiple realizations of binary facies model (sand and shale) across Low, Mid, and High heterogeneity scenarios defined by horizontal correlation range. A 2D dynamic reservoir model is developed using CMG-GEM, to simulate fluid behavior in the reservoir under CO2 injection scenarios. Numerical simulations are run for each geological heterogeneity realization and scenario to analyze fluid flow, CO2 migration and methane recovery under varying heterogeneity conditions. The simulation results reveal that increased heterogeneity accelerates CO2 breakthrough and amplifies uncertainty in CH4 recovery. Higher heterogeneity results in stronger pressure compartmentalization due to low-permeability shale barriers and enhances CO2 solubility trapping due to prolonged CO2-brine contact.

  • Open Access

    Where Will They Drill? Random Survival Forests for SpatiallyDependent Event Prediction

    by Dursun Dashdamirov, Dimitrios Belivanis, Jeff Stewart, Peidong Zhao, John Foster, Michael J. Pyrcz

    Forecasting drilling activity in unconventional reservoirs is essential for infrastructure plan-ning, logistical coordination, and regional economic management. While operators maintaininternal schedules, predicting activity remains challenging for external stakeholders due to thenonpublic nature of corporate strategies and limited access to proprietary geological data. Thiswork addresses the challenge by framing drilling expansion as a spatiotemporal time-to-eventprediction problem. We propose an end-to-end framework that integrates spatial dependencieswith survival analysis to estimate the cumulative probability of drilling over a multiyear horizon.Spatial and temporal feature aggregation scales are derived directly from indicator variogramanalysis, and engineered features explicitly encode spatial dependence, such as distance to thedrilling front and local angular coverage. Validated on fifteen years of drilling history acrossmore than 33,000 Eagle Ford Shale wells, the proposed model achieves a concordance index of0.77. Beyond this case study, the methodology can be applied to other high-dimensional spatialtime-to-event problems in subsurface engineering, as well as domains such as urban expansionand environmental hazard propagation.

  • Open Access

    Quantifying Uncertainty in Permeability Estimation Using Deep Learning and Generative Models

    by Oriyomi Raheem, Misael M. Morales, Michael Pyrcz, Carlos Torres-Verdín, Wen Pan, Yuanjun Li, Xiaohui Xiao, Rafael Centeno, Jay Chen, Pandu Devarakota

    Uncertainty quantification of well-log interpretation is essential to derisking subsurface exploration and development decision-making by providing possible scenarios for reservoir property distribution, fluid flow behaviors, and hydrocarbon potential. Well-log interpretation offers crucial insights into permeability variations, reservoir compartmentalization, mineral composition, and fluid mobility. Inherent uncertainties, such as those arising from geological heterogeneity, limited sampling, and non-uniform distribution of rock properties, can lead to inaccuracies that compromise petrophysical interpretation and formation evaluation. However, traditional data-driven well-log interpretation methods, which map well logs to formation properties based on core measurements, are primarily deterministic and fail to quantify uncertainty accurately. By leveraging deep learning and generative models, we introduce a probabilistic approach that significantly improves permeability estimation and uncertainty quantification. Our methodology integrates co-kriging techniques with Conditional Generative Adversarial Networks (cGANs) and Conditional Variational Autoencoders (cVAEs), establishing a quantitative relationship between kriged core, well-log data and permeability. Our approach enhances petrophysical property uncertainty estimations based on geostatistics by establishing a quantitative relationship between kriged estimates and flow-related properties. Training features are constructed using collocated co-kriging, capturing the cross-correlation between well logs (input features) and core data (output formation properties). Core bulk density, calculated from grain density, is kriged to well-log resolution to enable porosity estimation, while permeability is similarly kriged. A low-pass filter is then applied to smooth the kriged core bulk density, permeability, and estimated porosity, ensuring more accurate interpretations. The results reveal that cGANs and cVAEs consistently produce lower uncertainty estimates compared to traditional machine learning models. High-permeability zones exhibit lower uncertainty (approximately 3–5%), while low-permeability zones show higher uncertainty (10–15%). Traditional deep learning models tend to overestimate uncertainty, whereas generative models provide more reliable estimates. Additionally, applying kriged permeability data improves uncertainty estimations, further reducing uncertainty to 3% in high-permeability zones and 10% in low-permeability zones. To ensure broad applicability, the methods were tested on datasets from both carbonate and clastic reservoirs. In carbonate formations, prior classification steps are necessary to achieve accurate permeability predictions. The interpretation workflow improves permeability estimation accuracy and enhances uncertainty quantification across conventional and unconventional reservoirs. Additionally, this method is adaptable for CO2 injection and H2 storage wells, demonstrating versatility across various reservoir types.

  • Multiscale Joint Deformation–BHP Data Assimilation for CO 2 Storage Reservoir Characterization Using ES-MDA

    by Eunsil Park, Hyunmin Kim, Honggeun Jo, Michael J. Pyrcz

    Reliable prediction of coupled hydraulic–geomechanical behavior is essential for safe, long-term geological carbon storage (GCS). However, limited and spatially sparse observations from injection sites create large uncertainties in reservoir properties, reducing the reliability of CO₂ migration forecasts and monitoring, measurement, and verification programs. We present an integrated thermal–hydrological–mechanical data-assimilation framework using the Ensemble Smoother with Multiple Data Assimilation (ES-MDA) to jointly assimilate multi-physics and multi-scale observations. The framework combines time-series bottom-hole pressure (BHP) and spatial surface deformation data to constrain key hydraulic and geomechanical parameters in a fully coupled manner. We demonstrate the workflow in a 3-D channelized anticlinal aquifer model after performing sensitivity analysis, revealing updates to key reservoir parameters, including (1) hydraulic, (2) geomechanical, (3) wellbore design, and (4) reservoir/aquifer properties. BHP-only assimilation constrains permeability, well-skin factors, and sandstone relative-permeability parameters, whereas joint BHP–deformation assimilation further sharpens these flow-related estimates and improves the estimation of aquifer thickness and elastic properties, notably enhancing the centering of posterior Young’s moduli in sandstone and shaly sandstone. Joint assimilation also yields narrower posterior ensembles and lower mean squared errors during both the three-year assimilation period and the subsequent two-year forecast, enabling more reliable predictions of future reservoir pressure and surface deformation. This study shows that critical reservoir properties, including saturation-dependent relative-permeability functions and the Young’s modulus of the sandstone layer, can be robustly quantified by integrating both BHP and deformation data. The proposed multi-scale ES-MDA framework reduces predictive uncertainty, improves reservoir characterization, and supports secure, large-scale deployment of GCS.


  • Open Access

    Bayesian Updating with Multivariate Machine Learning-based Prior and Spatial Geostatistics-based Likelihood for Posterior Predictions of Unconventional Reservoir Performance

    by Dinghan Wang, John Foster, Yingda Lu, Andrea Thompson, Michael J. Pyrcz

    Accurate forecasting of hydrocarbon production in unconventional reservoirs requires integrating spatial heterogeneity with well-scale geological and engineering data. Traditional machine learning regression approaches often neglect spatial context, including spatial continuity, spatial trends and data locations, resulting in potential overestimation of uncertainty. To address this limitation, we introduce a Bayesian updating workflow with a multivariate machine learning model as a prior model and a spatial geostatistical model of well production as a likelihood model. The workflow is applied after normal score transformation of all features in Gaussian space to provide complete local uncertainty models from geostatistical kriging estimates and variances and to facilitate the use of closed-form analytical Bayesian updating. Our proposed workflow is demonstrated with 9 geological and 6 completion secondary features to forecast cumulative one-year oil production as the primary feature for an unconventional reservoir. A multivariate machine learning-based prior model is first calculated using local ridge regression with hyperparameter tuning to relate the secondary features to production. The likelihood model is calculated by ordinary kriging based on an additive deterministic trend and residual estimation. Bayesian updating with K-fold cross-validation is then applied to tune the workflow, demonstrating a reduction of the high uncertainty of the prior model, yielding a posterior distribution that is both more accurate and precise than the machine learning or geostatistical distributions on their own.

  • Open Access

    Gen-PINNs: Generative Adversarial Physics Informed Neural Networks for solving partial differential equations

    by Muhammad Muneeb Akmal, Kamy Sepehrnoori, Michael J. Pyrcz

    Physics Informed Neural Networks (PINNs), have been the standard data-free method for solving Partial Differential Equations (PDEs), using machine learning. With recent advances in Generative Adversarial Networks (GANs), many apply them for solving complex data-driven problems. However, the use of PINNs in GAN-based approaches is limited. In this work, we first identify limitations in the standard PINNs method for solving PDEs. Next, we combine and test multiple PINN extensions to overcome these limitations in a GAN architecture and then construct a novel Generative Adversarial Physics Informed Neural Network (Gen-PINN) framework to solve sharp- or shock-front bearing PDEs. In this approach, the generator learns the underlying PDE solution using dynamically weighted different loss components while the discriminator discriminates between the real known solution, at initial, boundary, and collocation points, against the generator output. Furthermore, several PINNs extensions are explored and combined inside the model to leverage the unique solution-finding ability of each method. Aspects are investigated, including: The objective loss functions of the generator and discriminator are specified by the type of loss and the loss schedule for the generator and discriminator, a variational form of mean square loss function, optimization algorithms, dynamic loss weighting, and Fourier input embeddings. The various methods are incorporated in the Gen-PINNs method and tested against standard PINNs to demonstrate effectiveness in using a synergistic unified approach to solve non-linear and higher-order PDEs. The improvements are demonstrated by testing complex PDEs, including Burgers’, advection-diffusion, Allen-Cahn, and the Kuramoto-Sivashinsky PDEs. The approach highlights the potential of improving shock-front PDE modeling by combining PINN extensions using Gen-PINNs.

  • Open Access

    A Diagnostic Method for Spatiotemporal Analysis of the Impact of Subsurface Reservoir Uncertainty on Dynamic Response Using Shapley values

    by Suin Choi, Jose Luis Hernandez Mejia, Honggeun Jo, Michael J. Pyrcz

    We propose a novel Shapley-value based diagnostic method to quantitatively evaluate how static geological model uncertainty influences spatiotemporal variations in dynamic reservoir responses. While conventional sensitivity analysis often fails to capture complex non-linear interactions among geological properties, we directly apply game theory-based Shapley values to numerical simulation results to isolate the intrinsic contribution of each property. We focus on three critical geological properties: sand fraction, porosity, and the Dykstra-Parsons coefficient, representing the degree of heterogeneity. By calculating Shapley values, the method quantifies the contribution of each geological property to dynamic reservoir responses across space and time, enabling diagnosis of shifts in property dominance throughout the production life cycle. Furthermore, the versatility of the proposed diagnostic method is demonstrated under an operational condition change, such as waterflooding, and different spatial correlations of reservoir properties. The method formalizes a set of spatiotemporal diagnostic questions by identifying which geological factor dominates, for example oil saturation, at a specific time, where this dominance is spatially localized, and how it evolves over time. This diagnostic perspective offers new insight into the mechanisms governing reservoir performance and supports informed reservoir management under geological uncertainty.

  • Open Access

    Gen-PINNs: Generative Adversarial Physics Informed Neural Networks for solving partial differential equations

    by Muhammad Muneeb Akmal, Kamy Sepehrnoori, Michael J. Pyrcz

    Physics Informed Neural Networks (PINNs), have been the standard data-free method for solving Partial Differential Equations (PDEs), using machine learning. With recent advances in Generative Adversarial Networks (GANs), many apply them for solving complex data-driven problems. However, the use of PINNs in GAN-based approaches is limited. In this work, we first identify limitations in the standard PINNs method for solving PDEs. Next, we combine and test multiple PINN extensions to overcome these limitations in a GAN architecture and then construct a novel Generative Adversarial Physics Informed Neural Network (Gen-PINN) framework to solve sharp- or shock-front bearing PDEs. In this approach, the generator learns the underlying PDE solution using dynamically weighted different loss components while the discriminator discriminates between the real known solution, at initial, boundary, and collocation points, against the generator output. Furthermore, several PINNs extensions are explored and combined inside the model to leverage the unique solution-finding ability of each method. Aspects are investigated, including: The objective loss functions of the generator and discriminator are specified by the type of loss and the loss schedule for the generator and discriminator, a variational form of mean square loss function, optimization algorithms, dynamic loss weighting, and Fourier input embeddings. The various methods are incorporated in the Gen-PINNs method and tested against standard PINNs to demonstrate effectiveness in using a synergistic unified approach to solve non-linear and higher-order PDEs. The improvements are demonstrated by testing complex PDEs, including Burgers’, advection-diffusion, Allen-Cahn, and the Kuramoto-Sivashinsky PDEs. The approach highlights the potential of improving shock-front PDE modeling by combining PINN extensions using Gen-PINNs.

  • Open Access

    Checking Generative Artificial Intelligence-Based Subsurface Models

    by Ahmed Merzoug, Michael J. Pyrcz

    This book chapter provides a comprehensive overview of generative artificial intelligence (genAI) applications in subsurface modeling. It introduces a suite of quantitative checks of genAI-based realizations. genAI for subsurface modeling is proposed as an alternative solution to traditional geostatistical workflows because it integrates several sources of information. We detail a minimum acceptance protocol that spans multiple checks of model evaluation: global univariate distribution (examined through histogram overlap, Q-Q plot divergence, Earth Mover’s Distance for continuous data, and total variation distance for categorical data); local uncertainty (assessed via cell-wise standard deviation and Shannon entropy); global uncertainty (evaluated using dimensionality reduction techniques coupled with overlap coefficients); and global spatial continuity (measured with variogram analyses). Additionally, we address local exactitude through conditioning metrics, identify conditioning artifacts, and check dynamic response using flow simulation recovery factors. The checks also include a stability analysis of the latent space via the Structural Similarity Index Measure (SSIM). These quantitative checks aim to enhance the reliability and applicability of genAI subsurface models in critical decision-making processes for resource development. This work is a practical guide for researchers and practitioners seeking to implement robust, uncertainty-aware genAI methodologies in subsurface modeling workflows.


  • Open Access

    GeostatsPy: Open-Source Geostatistics in Python

    by Michael J. Pyrcz

    GSLIB: Geostatistical Library by Clayton V. Deutsch and Andre G. Journel is the original spatial data analytics, geostatistics, open-source library built in FORTRAN with a function-based implementation to maximize workflow construction ease and flexibility. From simple toy problems for education to complicated subsurface model workflows, GSLIB is up to the task. Yet, it is difficult to teach the next generation with FORTRAN executables and PostScript visualizations. While there are a variety of efforts to add geostatistical methods to Python, I failed to find a package to meet my pedagogical needs in the modern Python language. I was compelled to reimplement GSLIB, function-by-function, often the nights before the associated lectures were given, to support my students. For reliability, I committed to rely only on the most common Python packages, such as NumPy, Pandas, SciPy and Numba. Yes, I took shortcuts, the methods are generally only available for 2D and there are missed opportunities to leverage existing code and to further accelerate for faster run times. The good news, it’s an open-source project, so if you see an opportunity to contribute you are most welcome. Participating in this project further expanded my respect for the vision and contributions of the original authors, Professors Clayton V. Deutsch and Andre G. Journel.