Bayesian modelling of sulphate isotopic composition in pristine, contaminated, and experimental environments for investigating microbial bacterial reduction

Abstract

The study encompasses three cases: a segment of the pristine carbonate aquifer with relatively low dissolved sulphate concentrations, often below 20 mg/L; a chalk aquifer contaminated by hydrocarbons, where microbial sulphate reduction was confirmed; and an experiment involving dissimilatory sulphate reduction. Bayesian models were developed using Python and its libraries, alongside the open-source probabilistic programming framework PYMC3, to assess the sulphate reduction hypothesis, ascertain the initial conditions of these processes, and demonstrate the applicability of this probabilistic method in elucidating hydrogeochemical processes, surpassing the widely used Rayleigh distillation model. The model produced robust and dependable results for both aquifer cases. Additionally, a model was devised to validate controllable and known conditions. The findings hold promise and may find relevance in similar hydrogeochemical processes. This study underscores the effectiveness of Bayesian modelling in revealing initial process conditions, particularly in situations where knowledge and data are limited. This limitation is often encountered with isotopic data, which remain relatively scarce due to the considerable effort required for material preparation and the costs associated with such investigations. Moreover, this study contributes to the identification of redox zonation, which holds significant implications for water quality-related concerns such as pollutant migration, remediation of hydrocarbon-contaminated sites, and meeting potable water quality standards regarding metal concentrations.