Characterizing the Effect of Different Sources of Uncertainty on Surrogates of Groundwater Models

Abstract:

The main sources of uncertainty in groundwater models are related to the hydraulic properties of the groundwater system, the driving forces, such as recharge and to the conceptualisation and implementation of appropriate boundary conditions.Characterizing the relative importance of each source of uncertainty in the input data and model structure can require an intractable number of model runs. This computational load may be reduced by employing surrogate models - simpler approximations of more complex codes.
Surrogate performance depends on the nature of the surrogate and which complex model runs are used to calibrate the surrogate. It tends to diminish away from these complex model snapshots.
In this paper we investigate how robust a Polynomial Chaos surrogate of a typical groundwater model is under common sources of uncertainty. We aim to provide guidelines as to when to re-calibrate a groundwater model surrogate, and to identify which sources of uncertainty tend to be more significant.