Proposals for unconventional gas developments typically involve predictive modelling of potential impacts on a groundwater flow system. The hydraulic properties of aquitards represented in such models are typically poorly constrained. The present study sought to demonstrate to what degree, if any, improved estimates of aquitard properties altered the sensitivity and uncertainty of predictions produced by a groundwater flow model.
This study used an existing groundwater flow model of coal seam gas extraction in an eastern Australian basin as the basis for the quantification of prediction uncertainty. Predictions generated by the model included the magnitude and timing of maximum drawdown in a confined aquifer. Laboratory analyses of core porosity–permeability relationships were combined with downhole neutron-density logs and upscaled using analytical and numerical methods to derive an improved basis for identifying the uncertainty of aquitard vertical hydraulic conductivity (KV) values. Monte Carlo sampling from specified hydraulic parameter prior distributions was used to estimate the uncertainty of modelled predictions before and after improved characterisation of aquitard KV values. Global sensitivity analysis metrics were used to assess the sensitivity of predictions to a range of hydraulic properties, prior to and following improved aquitard KV characterisation.
The inclusion of improved aquitard properties resulted in reductions in uncertainty for three of the four predictions. These predictions were sensitive to the vertical hydraulic conductivity of one of the two aquitards represented in the model. Conversely, the uncertainty of predictions of the spatial extent of drawdown was increased after the revision of aquitard properties.
More generally, this study serves as a demonstration of the use of Monte Carlo methods for the assessment of prediction uncertainty. In addition, global sensitivity analysis metrics can be used to comprehensively identify key relationships between predictions and model parameters. Such relationships may then be used to guide future data collection.