Lessons learned from model validation - a regulatory perspective

The Atomic Energy Control Board, the regulatory authority of the nuclear industry in Canada, has undertaken a project to validate the computer codes Femwater/ Femwaste. The strategy adopted to perform the validation was to calibrate the model based on historic monitoring data, predict subsequent behavior and then confirm the 1989 position of contaminant plumes from a uranium tailings impoundment. The existence of a large body of published data from the site (one of the most thoroughly studied uranium tailings sites in Canada) did not provide an adequate calibration of the site model. An initial field trip was necessary to collect additional calibration data. The sources of discrepancy between the measured and predicted positions of the contaminant plumes in 1989 are identified and quantified. The reliability of the predictions is demonstrated to be equal to the accuracy (bias plus uncertainty) of the model calibration. Recommendations for improvements in the validation strategy are made, including the need to begin the validation with an unambiguous problem definition (objective, variables to be assessed and validation criteria). 5 figs., 9 refs