Uncertainty Analysis - using fully- and extra-probabilistic approaches

Uncertainty analysis is an unavoidable task of risk assessments either for natural hazards like landslides, earthquakes, floods, volcanoes, etc., or for environmental issues like groundwater or soil contamination. When dealing with uncertainties, two categories should be considered as outlined by several authors:
1) “aleatoric uncertainty” (also named “randomness” or “intrinsic variability”) and arises from the natural variability owing to either heterogeneity or to the random character of natural processes (i.e. stochasticity). A common example of aleatoric uncertainty is the variability in weather.
2) “epistemic uncertainty” and arises when dealing with “partial ignorance” i.e. when facing “vague, incomplete or imprecise information” such as limited databases and observations or “imperfect” modelling.

Although the probabilistic setting has been used in a broad range of different applications, the use of probabilities as a tool to represent epistemic uncertainties has often been criticized in situations where the available data are imprecise, scarce, incomplete, vague, qualitative, etc. In such highly uncertain situations, the challenge is to formulate appropriate mathematical tools and models in a quantitative manner, on the one hand, accounting for all data and pieces of information, but, on the other hand, without introducing unwarranted assumptions. Therefore, to overcome the shortcomings of the pure probabilistic setting, several alternative representation methods have been developed: probability boxes, possibility distributions, Dempster-Shafer structures, etc.

The purpose of the short course is to describe how these new tools can be used to handle epistemic uncertainty for the different stages:
- Uncertainty representation;
- Propagation;
- Sensitivity analysis;
- Support for decision-making.

Comparisons with fully probabilistic approaches will also be performed. The short course will be supported by real cases taken from risk assessment studies for earthquakes (Rohmer & Baudrit, Nat. Haz., 2011), sea level rise (Le Cozannet et al., ERL, 2017), and groundwater contamination (Baudrit et al., J. Cont. Hydrology, 2007). These illustrations will be performed using R package “HYRISK” (https://cran.r-project.org/web/packages/HYRISK/index.html).