Some Limitations of Statistical and Probabilistic Models and Current Developments

Nuclear power plants and dams must demonstrate their safety against events with a return period of 10,000 years. However, determining such rare events from observations alone is difficult because time series often cover only years or decades, and in some very rare cases, centuries. Various statistical techniques attempt to overcome the limitations of extrapolating such extreme return periods. In some cases, pre-instrumental data can be used, such as paleo-seismic or paleo-hydraulic data. In recent years, a new approach has emerged that promises to cover phenomena that cannot be detected in the data because of the limitations of the data series. These are model-based methods. Especially in Switzerland, this approach has been used for example in the EXAR project for precipitation and floods, but also in other recent studies on extreme precipitation and temperatures. This article aims to highlight the opportunities, but also the risks, of this approach. The importance of such considerations is that this approach is not only found in the field of climatology, but also in the field of structural safety or earthquake engineering.