PNNL

Abstract

Mineral trapping is pursued as a mechanism for geological CO2 sequestration (GCS) as it per-manently stores CO2 in solid phases or minerals. However, CO2 mineral-trapping mechanisms are poorly understood due to (1) lack of su?cient ?eld and laboratory data characterizing this complex processes, and (2) challenges to develop site-speci?c reactive-transport models coupling ?uid ?ow and geochemical reactions occurring at various temporal (from milliseconds to years) and spatial (from pore (millimeters) to ?eld (kilometers)) scales . Reactive transport with ad-ditional complexities such as heterogeneity can make the simulation outputs even more di?cult to interpret because of complex nonlinearity and multi-scale interdependencies. Furthermore, the values of model outputs such as concentrations can vary by several orders of magnitude, making it harder to correlate and characterize the impact of the variables via traditional data interpretation techniques such as exploratory data analyses. Recently, machine learning (ML) has shown promise in feature discovery and in highlighting hidden mechanisms, that cannot be obtained by existing data-analytics and statistical methods. In this study, we applied an unsupervised ML approach, non-negative matrix factorization with custom k-means clustering (NMFk) to the data generated by reactive-transport simulations of geologic CO2 sequestration. The reactive-transport data con-sisted of 19 attributes including four physio-chemical variables (pH, porosity, aqueous CO2, and sequestered CO2), six chemical species (K+, Na+, HCO3-, Ca2+, Mg2+, Fe2+), and nine minerals (calcite, dolomite, siderite, ankerite, albite, illite, clinochlore, kaolinite, and smectite), over a pe-riod of 200 years of simulation time. The simulation data used was for Morrow B sandstone at the Farnsworth hydrocarbon unit in Texas. Data are sampled at two locations with the model domain:(1) at the injection well and (2) 200 m west of the injection well. The injection was performed for a period of 10 years. Using NMFk, we estimated the temporal interdependencies among the 19 attributes over a span of 200 years. We found that NMFk was able to identify four reaction stages and their dominant attributes; these cannot be directly discerned through traditional visualization (e.g., line plots, Pareto analysis, Glyph-based visualization methods) or exploratory data analysis tools of the simulation data. The four stages were: reactions in the injection phase followed by short-, mid-, and long-term reactions. The NMFk analysis also revealed that 10 attributes are dominant among the 19 attributes. These dominant attributes for mineral trapping include cal-cite, dolomite/siderite, clinochlore, kaolinite, Na+, K+, Ca2+, Mg2+, pH, and aq. CO2. Finally, at late times, our results showed that calcite plays a major role in mineral trapping with insigni?-cant contribution from siderite, ankerite, and clay minerals. All these ?ndings make the proposed unsupervised ML-model attractive for reactive-transport sensing towards real-time GCS monitor-ing under limited or no labels. Moreover, NMFk can e?ciently discover and identify important chemical species for informative and cost-e?ective data sampling strategies during a ?eld campaign.