A decision‐making framework for flood risk management based on a Bayesian Influence Diagram

Abstract

We develop a Bayesian Influence Diagram (ID) approach for risk‐based decision‐ making in flood management. We show that it is a flexible decision‐making tool to assess flood risk in a non‐stationary environment and with an ability to test different adaptation measures in order to agree on the best combination of adaptation measures and the best time to invest in flood adaptation. IDs use Bayesian statistics which apply prior probabilities to produce posterior probabilities and, hence, use Bayesian probabilistic thinking to describe relationships between variables in a system. . Hence, we allow for assessing the risk of something we ?believe? may occur in the future. An ID has two layers; 1) a graphical description of the system built up by system variables, adaptation measures, costs/benefits of these measures and the dependencies of all these, which is an effective means to communicate the system configuration, and 2) conditional probability tables (CPTs) in which the domain of all possible states taken by the variable is listed combined with conditional probabilities of any state of that variable. When the ID is compiled, i.e. posterior probabilities are calculated; the network can be updated each time new values of variables are observed, assuring that the risk assessment is constantly based on best available knowledge for each variable. Input data to IDs can come from multiple sources, and since each variable is described with a probability density function (pdf) this method provides an effective means to describe uncertainty in the system. Hence, an ID contributes with several advantages in risk assessment and decision‐making. We present an ID approach for risk‐ based decision‐making in which we improve conventional flood risk assessments by including several types of hazards into the assessment. By doing so, we explicitly consider the risk from concurrent events. Further, we add large scale weather patterns to the risk assessment as an additional variable to describe the occurrence of extremes. This allows using projected changes in large scale circulations from climate models to estimate pdfs of extreme events in a future climate. Our method provides means to assess non‐ stationarity of flood risk by including several time steps in the risk assessment (? ström et al., 2013). Hence, our approach effectively communicates to decisionmakers how risk changes over time and the uncertainty related to these changes. We combine a flexible impact assessment method with our ID that can assess the overall risk in a given area as well as within subareas. This impact assessment provides for a transparent and robust assessment of both instant and long‐term benefits for different adaptation measures and combinations of these. Adaptation options can be tested at different points in time (in different time slices) which allows for finding the optimal time to invest. The usefulness of our decision‐making framework was exemplified through case studies in Aarhus and Copenhagen. Risk‐based decision‐making is difficult, and considering the partly unknown processes related to anthropogenic climate change we need to model a very complex system. In our study we showed that IDs are a noteworthy alternative as decision‐making method in flood risk management and is a useful method when several hazards and their simultaneous occurrence need to be assessed. The approach provided several benefits such as a transparent explanation of the system at risk, clear description of the uncertainty in the system and the changes over time, and flexible means to assess the best combination of adaptation measures.