Coal Seam Gas production in the coming decades will involve disposal of large amounts of co-produced water into deep aquifers. Prior to the implementation of larger-scale injection schemes, it is important to perform field injection trials that generate sufficiently meaningful data sets to allow for the assessment of both the hydrological and geochemical impacts on the target aquifer. The present work illustrates the use of reactive transport modelling for data analysis from a field experiment where arsenic mobilisation was observed. This type of modelling study requires integration of multi-scale data and models with a relatively high parametric uncertainty. The parametric uncertainty, however, can lead to significant predictive uncertainty when the field-scale reactive transport model (RTM) is employed for the prediction of the long-term groundwater quality evolution.
Arsenic sorption behaviour was studied through laboratory experiments and modelled using a surface complexation approach. A field-scale RTM that incorporated the laboratory-derived model was used to simulate the injection trial and to predict the long-term fate of arsenic. Here, we propose a new practical procedure for better integration of laboratory and field-scale models in order to quantify predictive uncertainty. The approach alleviates a significant proportion of the computational effort required for uncertainty quantification. The results illustrate that both desorption and pyrite oxidation have likely contributed to arsenic mobilisation observed during the trial. The predictive simulations show that arsenic levels are likely to remain very low if the potential for pyrite oxidation is minimised through a complete deoxygenation of the injectant. The proposed modelling and predictive uncertainty quantification approach can be implemented for a wide range of groundwater studies that investigate the risks of metal(loid) or radionuclide contamination.