| This dissertation presents a research on the stochastic inverse modelling of groundwater flow and mass transport. The sequential self-calibrating method, an approach that allows to generate multiple equally likely solutions to the inverse groundwater flow problem, has been extended to handle transient groundwater flow, 3-D groundwater flow, groundwater flow in fractured media and the coupled inverse modelling of both groundwater flow and mass transport. Furthermore, the impact of measurement errors has been studied. The algorithm has been implemented in the software INVERTO. The extended approach has been tested in numerous synthetic studies and real-world case studies. The real-world case studies are all related with the underground storage of nuclear waste.; The self-calibrating method was extended to handle transient groundwater flow and the joint conditioning of transmissivity and storativity fields. The adjoint state method is used to calculate in an efficient way the gradient of the objective function. The transient head data just yielded limited modifications in the transmissivity field in the WIPP-case study (New Mexico, U.S.A.), while the transmissivities of a single fracture in the Aspö-case study (Sweden) changed noticeably. The impact of the transient head data is related with the extent of the zone of the aquifer that is influenced by the transient events. In the Aspö-case study it was found that modifications in the transmissivity variogram parameters and the estimated transmissivity measurement error just had a limited influence on the final results. The impact of considering uncertainty in the boundary conditions was somewhat larger. In both case studies consistent estimates were obtained of a single storativity value.; Finally, the approach was extended to the coupled inverse modelling of groundwater flow and mass transport. The extension of the inverse modelling approach to transport required a numerical scheme to solve the mass transport equation, an expression for the objective function that also included the mismatch to the concentration data and a method to calculate the gradient of the objective function. (Abstract shortened by UMI.)... |
