Modelling anion-accessible porosity of rocks based on different approaches

Anionic radionuclides are relevant contributors to the overall dose that may originate from an underground repository for radioactive waste. Clays are important parts of engineered and natural barriers of repositories due to their sealing properties. As clay surfaces are negatively charged, anions are depleted in the pore space near the clay surfaces. This partial exclusion of anions strongly affects their transport. To make reliable predictions of the evolution of a repository, e.g., for safety assessments, a thorough understanding of this phenomenon is required, as well as the ability to model the exclusion effect for different conditions.

Unfortunately, the degree of anion exclusion depends on many parameters, including the mineralogical composition of the rock, the porosity, and the porewater chemistry. Moreover, it depends on rock properties difficult to quantify such as texture or generally the pore space architecture. While the basic principles behind anion exclusion are understood and various models exist, it is not straightforward to apply these models to different rock types or different chemical conditions. At the same time, the direct determination of the anion accessibility (an average property defined as the fraction of the pore space fully accessible to an anion) by diffusion experiments is very time consuming. Fortunately, a recent deep drilling campaign in Switzerland provided a large data set including both, rock properties and anion accessibilities.

Here we profit from this data set and compare different methods to derive average anion accessibilities for various rock types and conditions. In a first approach, we build a chemical-structural model for a rock based on the fractions of different phases (minerals, porewater), the porewater chemistry, and assumptions regarding the distribution of the porewater. In a second approach, we apply Machine Learning (ML) on a training data set and build a model based on the most influencing parameters, including clay-mineral content, water content, and porewater chemistry.

Both approaches are performing relatively well for clay-rich units. However, they show weaknesses in other lithologies, especially for rocks with very low water contents or with (presumably) very specific texture. For such rocks, there is a lack of knowledge to develop suitable chemical-structural models. Also, the data base with regard to anion exclusion is currently still small for such rocks, which clearly limits the ML approach. Thus, extending the appropriate knowledge, e.g., by microstructural investigations, as well as the corresponding data base are considered as promising next steps.