For this worked example, we rebuild an old USGS model. The focus of the original model was protection of water supply wells that serve a small town in New Hampshire. Those who built the original model did a good job – given the technology that was available at the time. We show how modern data assimilation and uncertainty analysis methods can support superior processing of groundwater datasets with little extra trouble.
We use MODFLOW-USG for simulation of groundwater movement, and mod-PATH3DU for particle tracking. Model calibration uses PEST_HP while uncertainty analysis uses PESTPP-IES. You can download all of our files, as well as a document that explains what we did.
This worked example raises some interesting issues. It shows the importance of working in a highly parameterized world. This allows the modelling process to extract information from field measurements by supporting a good fit of model outputs with those measurements. It also allows the modelling process to express the repercussions of information insufficiency by quantifying post-history-matching predictive uncertainty.
This worked example also shows the importance of identifying what data are worth respecting and what data are not. Contributing area (i.e. capture zone) predictions can be particularly sensitive to parameters that may compensate for model inadequacies, or dubious data, during history-matching. They can therefore suffer history-matching-induced bias. This is especially the case where a production well may draw some of its water from a nearby stream.
We calculate probabilistic capture zones twice – once with a three-layer model and once with a one-layer model. If near-stream data are scarce, or lacking in credibility, the one layer model can do just as good a job as the three layer model. This may ease the burden of contributing area analysis where a local authority is required to undertake many such studies for a suite of public water supply wells.