Sorption process is one of the most important mechanisms influencing the fate of contaminant (e.g., radionuclide) transportation in subsurface geologic systems. Due to this reason, quantifying the sorption process has become a principal component of contaminant transport modeling and several approaches have been adopted to numerically describe the sorption phenomenon: e.g., distribution coefficient (Kd) and surface complexation modeling (SCM). Kd approach is simple to implement, however, Kd values are not compatible in different environmental conditions. While the SCM approach is more versatile than the Kd approach, obtaining global consensus among reported SCM constructs is challenging since each SCM construct is based on divergent assumptions. In light of these limitations, we have developed a more comprehensive SCM construct by compiling raw sorption data reported from various literature sources. In addition, we also have employed a new hybrid machine learning (ML) model, named the LLNL Speciation Updated Random Forest (L-SURF), which quantifies sorption by implementing not only data-driven random forest decision making process but also geochemical aqueous speciation calculations. In this study, the newly developed approaches were examined for the sorption of two Se oxidation states (IV and VI) to several iron oxide mineral phases (ferrihydrite, goethite, etc.). In Se(VI)-ferrihydrite case, we mined 281 data points from seven different references and both comprehensive SCM construct and hybrid ML model reasonably predicted sorption phenomena with a weighted Pearson correlation coefficient value of 0.90 (for optimized SCM construct) and 0.92, respectively. Our work demonstrates the use of community data to provide global surface complexation constants which can capture raw sorption data from different sources. Additionally, we introduce a novel hybrid ML approach based on the same community mined dataset, creating a promising surface complexation modeling foundation for contaminant transport modeling.

Community data driven surface complexation modeling and hybrid machine learning approaches to predict selenium-iron oxide sorption

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.