Parameterization of Large Regional Models: The View from the Trenches

Parameter estimation–inferring parameter information through history matching and inverse methods–is often the only tenable way to populate properties in a large-scale regional model. In a Bayesian sense, the properties estimated in this way, and later used to make predictions, are conditional upon the data used to calibrate the model and all the assumptions made in the underlying model. This can result in a disconnect between parameters estimated at the laboratory, aquifer-test, and model scales. Additionally, more types of data–each with a different support volumes and level of information–are becoming available with more near-surface geophysical measurements being made on a regional scale. To take advantage of these data, a combination of flexible parameterization through distributed parameters, regularization, and weighting of various observation types can lead to defensible and useful results.

The USGS Groundwater Resources Program uses regional models and data assimilation to analyze and predict water availability in major regional aquifer systems in the United States. The inverse modeling process applied in several of these projects provides valuable lessons on the meaning of parameters at the large scale. Inverse modeling is used to estimate model parameters but—more importantly—to improve understanding of the conceptual framework the models are built on.