Uncertainty in hydrogeological modelling.

Abstract

Hydrogeological models are built to predict groundwater flow and the fate of contaminants in the subsurface. After the crucial step of building a conceptual model that includes the processes that should be accounted for, parameter values must be assigned to the components of the model. Measured values of these parameters are available only at a few locations, as is the case for transmissivity, hydraulic conductivity or porosity. Therefore, before making predictions about the movement of contaminants in the aquifer, it is necessary to predict the parameter values at unsampled locations. Given the spatial heterogeneity of the parameters involved, this prediction is always uncertain. Model parameter uncertainty propagates to flow-response variables and further to transport predictions. Parameter uncertainty can be modelled using stochastic methods. Stochastic simulation is used for the generation of alternative spatial realizations of the parameter values, which are then used as input to groundwater flow and mass transport models to obtain frequency distributions of the response variables, e.g. flow velocities, arrival times or concentration levels. These frequency distributions help in making risk-qualified decisions. In order to make these frequency distributions as precise and accurate as possible, it is necessary to incorporate all relevant information in the parameter uncertainty model, i.e. it is necessary to condition the parameter realizations to all direct and indirect information. With this aim, new techniques have recently been developed in hydrogeological modelling. One such technique, for the generation of conductivity realizations conditioned to conductivity, piezometric head and geophysical data, is the self-calibrated method.