Repositories have to isolate radioactive waste (radwaste) and some toxic wastes from the environment for hundreds to hundreds of thousands of years. For many scientists and engineers, and especially the general public, such time spans are beyond comprehension and, consequently, they have grave doubts as to the safety of any such waste repositories. That repository performance over these long time scales can only be assessed by the use of complex mathematical models (normally called performance assessment, or PA, models) only adds to the mistrust of many. How then can people be convinced that it is possible to assess the performance (and thus ensure the safety) of a repository over the long timescales of interest? One way is to address the robustness of the PA models, by clearly indicating the form and extent of model testing carried out within the repository PA. Not only can this show that the individual component parts of the complex structure which constitutes most PA models have been checked, but also that the ‘mathematical black boxes’ constitute an acceptabe representation of the repository system.

Part of the problem undoubtedly lies in the unusual nature of radwaste disposal: in most major engineering projects, such as bridge construction or aerospace engineering, the designs are tested against a range of laboratory experiments backed up by expert judgement based on experience with the same or similar systems. Here repository design deviates from standard engineering practice in that no high-level waste (and only a few low- and intermediate-level waste) repositories yet exist and, even when they do, testing their compliance to design limits will be somewhat difficult due to the time scales involved. In addition, the irrational fear of most things radioactive means that most people require some greater form of ‘proof’ that a repository is safe than they are willing to accept for other engineered systems. This being the case, significant additional effort must be expended within the radwaste industry to make it completely clear that the PA models can adequately predict the long-term behaviour of a repository.

Traditionally, PA modellers have placed much weight on laboratory data for the construction and testing of PA models and, with only a few exceptions, have not integrated in their PA reports data from natural analogues and in situ experiments. The over-dependence on laboratory data is understandable in that the information is produced under well understood, fully controlled conditions and thus the modellers feel they can place a high degree of confidence in the results obtained. Unfortunately, the full complexity of a repository cannot be re-created in a laboratory and it is necessary to address processes which are influenced by natural heterogeneities, which include large degrees of uncertainty and which operate over very long timescales. In this case, it is necessary to supplement laboratory data with information from in situ field experiments and natural analogues.

The advantage of natural analogues over short-term laboratory experiments is that they enable study of repository-like systems which have evolved over the geological timescales of relevance to a radwaste repository safety assessment (rather than the days to months usual in laboratory tests). However, by their very nature, natural analogues often have ill-defined boundary conditions which may be better assessed under the well constrained (if less relevant) conditions of a laboratory. Well designed, realistic in situ field experiments can bridge the gap between the laboratory and natural analogues by offering repository relevant natural conditions with some of the constraints of the laboratory (and intermediate timescales). In short, combining information from the three sources (long-term and realistic, if poorly defined, natural analogues, medium-term, better constrained, in situ field experiments and short-term, less realistic but well defined laboratory studies) can provide greater confidence in the extrapolation of laboratory derived data to repository relevant timescales and conditions.

This paper will concentrate on presenting a model testing scheme which both promotes transparency (for the sake of technical peer reviewers and the general public alike) and which aims at a thorough test of PA models. In addition, several recently published PAs will be critically examined and the form, extent and transparency of testing will be discussed with a view to improving confidence in the robustness of the models and thus the perceived safety of specific radwaste repository designs.