Shall We Care about Optimal Statistical Decision Theory When Designing a Flood Protection Dike under Climate Variability?
Jacques Bernier1, Eric Parent1, Merlin Keller2, Kouloud Ghorbel2
1MORSE (Statistical Research Unit), AgroParisTech, Paris, France; 2Industrial Risk Management Unit, EDF, Chatou, France

The most important but perhaps the most inappropriate hypothesis to be made when designing a flood protection work is to assume a constant statistical behaviour of the dangerous phenomena to be protected from. However, in a changing environment, less can be learned from the past to predict the future. Yet, some assumptions of stationarity remain necessary for statistical analysis and probabilistic risk assessment. How should civil engineers change their standard practice to adapt to such a loss of (statistical) memory? How can we quantify the additional source of uncertainty stemming from climate variability?

We first recall the statistical decision theory for risk assessment, exemplified in the case of a flood protection dike. Second, when learning from the past becomes troublesome, recourse can be made to models mimicking sudden changes, such as the shifting level model. As a result, we show that more uncertainty is incorporated into the representation of the damaging events to come, yielding to a more cautious design and providing a higher value of information.

Keywords: Bayesian decision theory; Hydrology; Changing environment; Shifting Level model

Biography: I work as a Professor in applied statistics and probabilistic modeling for environmental engineering. My research group belongs to AgroParistech, an academic institution from the Ministry of Agriculture. The laboratory enrols PhD students and postgraduate engineers trained to work at the intersection between statistics, decision-making theory and environmental engineering. I co-authored three books (in French), one on Bayesian statistics for environmental engineering, the second on theoretical and algorithmic aspects of Bayesian theory and the third one depicts various cases with real data from various domains of applications. My broader interests include “Bayesian Statistics at work”, especially in case studies from various fields while working with my PhD students, under contract with industrial companies or public institutions