In groundwater modeling, uncertainty in parameters and predictions can be reduced by history-matching the model against observations of groundwater levels. When observations span different periods or come from pumping tests, complex calibration schemes may be required, potentially making optimization unfeasible. In such cases, the modeler may choose to lose information by reducing or simplifying the observation set in exchange for a manageable process.
This study employs the Ensemble Space Inversion (ENSI) methodology for history-matching a groundwater model of a complex site against a large number of observations. ENSI estimates "super parameters" instead of native model parameters, using an ensemble of random samples from prior parameter probability distributions. A key advantage of this method is that it requires far fewer super parameters than model parameters, reducing the number of model runs needed to calculate parameter sensitivities.
ENSI was applied to a site characterized by heterogeneous pyroclastic and volcanic deposits, with three distinct aquifers. Various hydraulic tests were conducted to determine site-specific hydrogeological parameters, assess vertical conductance, derive pumping well characteristics, and collect piezometric data. A six-layer numerical model was developed using MODFLOW-USG. Hydraulic parameters were calibrated via pilot points, with values interpolated through kriging.
Compared to history-matching the same model using a standard GLM method applied to model parameters, ENSI provided several advantages: fast convergence to low objective function values, significantly reduced execution time, and natural parameter field distributions.
Though ENSI does not perform uncertainty estimation, it offers a fast and effective way to obtain a single history-matched model, which could serve as a starting point for ensemble-based uncertainty assessment.