Our first worked example report describes design and development of a predictive model built to examine water supply security of a small town in South Eastern Queensland.
It shows how even with a relatively small observation dataset, sufficient information about hydraulic parameters, their distribution and uncertainty can be extracted during history matching. This is possible due when observations are used in conjunction with ‘Soft data’ embodying qualitative and intuitive knowledge of the behaviour of the groundwater system. This work also challenges the notion that model boundaries should be set at distance; instead dynamic boundaries driven by a simple recharge model are used and directly assessed in terms of their contribution to prediction uncertainty.
The example demonstrates application of PEST_HP for history matching with PESTPP-IES then being used to generate 250 history-match-constrained random parameter fields that are used to assess uncertainties in water supply security. Interestingly, uncertainties in future weather contribute more to prediction uncertainty than those from the history matched model, despite a small observation dataset.
Our second worked example will be ready in a few months’ time. Its context is the design of a dewatering system for an open cut mine. It will address the issue of connected permeability imposed by structural features such as faults. Sometimes these features are almost invisible to the history-matching process. Our work will explore the use of regularization methods that can purposefully look for these features, and of uncertainty analysis methods that can accommodate the possibility of their existence, even if their exact locations are unknown.