Date of Award

May 2026

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Petroleum Engineering

First Advisor

Hadi H. Jabbari

Abstract

Underground hydrogen storage (UHS) depends on complex interactions among reservoir rocks, brine, and gas under redox conditions induced by H2 injection. Although wettability is often treated as a static parameter, redox reactions can shift pH, aqueous speciation, and mineral surface charge, thereby altering contact angles and, consequently, injectivity, trapping, and recovery. This work tests the hypothesis that dominant redox pathways in saline aquifers and depleted reservoirs, specifically Fe(III) and sulfate reduction, systematically modify mineral surface stability and charge, producing predictable shifts in wettability. To address this, we developed a physics-guided machine learning workflow that integrates mechanistic geochemical understanding with empirical data. We use PHREEQC to perform scenario modeling and surface-complexation calculations for quartz, calcite, and representative clays, generating a suite of mechanistic descriptors (such as pH, pe, ionic strength, and surface-charge proxies) under realistic pressure–temperature–salinity (PTS) conditions. The novelty of this work lies in integrating PHREEQC-derived geochemical descriptors with curated experimental contact-angle data from the literature to develop a screening-level, physics-guided surrogate framework. We train interpretable ML models on these hybrid datasets and use grouped cross-validation to account for inter-laboratory variability. We attribute feature effects using SHAP, conditional permutation, and ALE to identify stable, defensible drivers of wettability. Findings include identification of chemical regimes that shift systems toward more water-wet or gas-wet behavior and a quantitative ranking of controlling variables. This research provides a novel tool for UHS site screening, offering guidance for cushion-gas selection, brine management, and optimization of hydrogen retrieval efficiency.

Download

Share

COinS