Efficient Model Calibration Using Submodels

Abstract

Groundwater models tend to become increasingly detailed to accommodate increasing data availability and higher accuracy demands from stakeholders. As runtimes increase almost quadratically with the number of model cells, this makes the models ever more computationally demanding. This high computational demand introduces challenges for the history-matching (calibration) process as this is an algorithmic process that needs hundreds or thousands of model-runs to obtain the model sensitivities needed to estimate parameters. Model runs may take hours or days to complete which in fact, is often a reason to discard the history-matching all together. As a solution, we present a practical approach to use sub-modeling in combination with parallelization for automatic history-matching. Therefore a large model is subdivided into smaller models to carry out the sensitivity simulations. With a realistic case the method is elaborated, after which the method is demonstrated in the history-matching of the transient Dutch National Groundwater Flow model. In this manner the model, which consisted of over 12 million model cells, could be optimized using 416 sub-models and altogether 2,188 parameters in 1 week. This would take years to complete in a conventional way.