Technical Report NTB 88-14

Inverse modeling of Chalk River blockHydrocoinLevel 2 / Case 3Level 3 / Case 5A

Within the framework of the international project HYDROCOIN, a block of fractured monzonitic gneiss within the facilities of Chalk River National Laboratories, Canada, was selected as a test case to study and develop strategies for the calibration and validation of groundwater flow models.

Adopting a quasi-three dimensional formulation, the fractures were simulated by two-dimensional finite elements and the rock mass was simulated by strings of line elements. The models were calibrated using, first, steady-state data and, second, transient data. Model calibration involved both identification of model parameters and model structure. Model parameters were obtained by automatic estimation based on measures of the model response and prior information about the model parameters. Excellent agreement between measured and computed heads was obtained for the transient runs. However, such match was only fair in steady-state. Model Structure Identification Criteria were used to rank the performance of several model structures. In the steady state the Model Structure Identification Criteria did not strongly support increasing the model complexity. However, it is also believed that the information content of the steady-state data was quite poor. In contrast, the transient data being both more numerous and more informative than steady-state data, allowed the Model Structure Identification Criteria to suggest more complex models. The validation runs were performed on data corresponding to interference pump tests different from the ones used for calibration. The prediction errors in these runs were relatively small and consistent with the calibration uncertainty. Furthermore, the ranking of the models performances during validation runs was the same as the one obtained at the calibration stage, using Model Structure Identification Criteria.