Numerical groundwater modelling to support mining decisions is often challenging and time consuming. Simulation of open pit mining for model calibration or prediction requires models that include, unsaturated flow, large magnitude hydraulic gradients, and often require transient simulations with time varying material properties and boundary conditions. This combination of factors typically results in models with long simulation times and/or some level of numerical instability. In modelling practice, this fact can result in reduced effort for predictive uncertainty analysis, and ultimately decrease the value of the modelling to support decisions. This study presents an early application of the iterative ensemble smother (IES) method of calibration constrained uncertainty analysis to address the challenges of mining models and uncertainty quantification. The IES method was applied with PEST++ IES software and facilitated by highly parallelized computing using the Amazon EC2 cloud computing service.
An operating open pit mine in South Australia required estimation of long-term recovery pit water levels and inflow rates to support decisions regarding the long-term environmental impact of the project, and the feasibility of a proposed pumped hydro energy storage system. A groundwater observation dataset was available consisting of static water level measurements taken prior to the most recent mining activity from both project specific observation bores and public databases at 98 locations. Transient observations of groundwater level changes over 7 years of mine development were available at 16 locations. Initial model simulations indicated that the application of traditional finite difference based methods of calibration and uncertainty analysis would be complicated by low magnitude numerical instabilities, and require excessive computational effort due to multi-point derivatives or highly refined model grids and long simulation times.
The IES calibration successfully produced 150 model parameter realizations that acceptably reproduced groundwater observations. The flexibility of the IES method allowed for the inclusion of 1,493 adjustable parameters and geostatistical realizations of hydraulic conductivity fields to be included in the analysis. The IES method out-performed finite difference methods when model simulations contained small magnitude numerical instabilities.